Operational Impacts of Responsive Electricity Loads:
A Modeling Framework Including Policy Implications for Cyprus
ARCHIVES
by
Matthew Bremer Bruchon
, S%CH4;6ZETTS INS'rWE
i OF TECHNOLOGY
B.S. Computer Engineering
B.S. Electrical Engineering
B.A. English
North Carolina State University, 2008
NOV 2 0 2013
UBRARIES
Submitted to the Engineering Systems Division
and
Department of Electrical Engineering and Computer Science
in partial fulfillment of the requirements for the degrees of
Master of Science in Technology and Policy
and
Master of Science in Electrical Engineering and Computer Science
@
at the
MASSACHUSETTS INSTITUTE OF TECHNOLOGY
September 2013
Massachusetts Institute of Technology 2013. All rights reserved.
A uthor ............
......................
Technology and Policy Program, Engineering Systems Division
Department of Electrical Engineering and Computer Science
August 9, 2013
C ertified by ..........
.......
1
...
...
.. ...
.. ...
..
Stephen Connors
Director, Analysis Group for Regional Energy Alternatives, MIT Energy Initiative
Thesis Supervisor
C ertified by .............
........................
James L. Kirtley, Jr.
Professor of Electrical Engineering and Computer Science
Thesis Reader
Accepted by ...........
A ccepted by ............
............................
Dava Newman
Professor of Aeronautics and Astronautics and Engineering Systems
Director, Technology and Policy Program
.
...........................
.
(Leslie
(
A. Kolodziejski
Professor of Electrical Engineering
Chair, Department Committee on Graduate Students
Operational Impacts of Responsive Electricity Loads:
A Modeling Framework Including Policy Implications for Cyprus
by
Matthew Bremer Bruchon
Submitted to the Engineering Systems Division
and
Department of Electrical Engineering and Computer Science
on August 9, 2013, in partial fulfillment of the
requirements for the degrees of
Master of Science in Technology and Policy
and
Master of Science in Electrical Engineering and Computer Science
Abstract
In order to meet EU mandates, the island nation of Cyprus must raise penetration of renewable energy from roughly 5% in 2013 to 16% in 2020. This means Cyprus will need
economical ways of balancing intermittency, a special challenge for small island power systems which have less inertia, narrower reserve margins, and high fuel costs for thermal
generators.
This thesis explores the potential of demand response programs to help integrate renewables in Cyprus from an hourly unit commitment perspective. A stochastic optimization
model of the nation's power grid is presented, including thermal generators, wind, solar photovoltaic, and concentrated solar power with thermal storage. Demand response programs
are modeled as a variety of shiftable or curtailable loads, with configurable parameters
such as: energy capacity, maximum operation time of a load, maximum time a load can
be shifted, lead time required to shift a load, and minimum interval between calls to shift
a given consumer's load. The model includes loads from the residential, commercial and
desalination sectors.
The model is run on scenarios with and without a planned transition from fuel oil
to natural gas generation, with either a gradual or a fast economic growth between 2013
and 2020. In all scenarios, the model finds that demand response can help Cyprus reduce
electricity costs, harvest more useable energy from wind and concentrating solar power, and
reduce carbon dioxide emissions.
Thesis Supervisor: Stephen Connors
Title: Director, Analysis Group for Regional Energy Alternatives, MIT Energy Initiative
Acknowledgments
I appreciate the mentorship of the faculty and researchers I have worked with at MIT.
Stephen Connors, my advisor, offered guidance at every stage of the project, improved my
understanding of energy topics, and made it a great (and enjoyable) learning experience
along the way. David Marks helped lead the Cyprus research effort from its formation and
provided me with generous financial support from the Cyprus Institute and the Klegerman
Fellowship. James Kirtley helped review this work and gave useful feedback for improving
it. Daniel Livengood introduced me to this project, helped formulate a research topic, and
assisted with model debugging; his contributions were essential.
Luciana Herman's course at the Harvard Kennedy School taught me about policy writing, and her feedback helped shape the first drafts of this work into a readable form.
D. Karl Critz's previous work is the basis for much of this model, including code he
originally wrote which is leveraged heavily and extended here.
Alexandros Charalambides at the Cyprus University of Technology provided solar resource data and Constantinos Varnavas at the Electricity Authority of Cyprus provided
power system data, making this model more realistic.
Jackie Donoghue's administrative support, as well as the advising of Barbara DeLaBarre,
Krista Featherstone and Janet Fischer, kept my research project running smoothly.
My student colleagues have helped this project in various ways, including by proofreading my work, lending computing resources, explaining the Cyprus context, translating Greek
sources, and assisting with thesis logistics. Those colleagues include (but are not limited to)
Claudio Vergara, Karen Tapia-Ahumada, Dan Cross-Call, Amy Rose, Joanna Karkatsouli,
Christina Karapataki, Leebong Lee, Sandra Jenkins, Nathan Lee, and Woei Ling Leow.
My family has helped me in countless ways to get where I am today.
To all of these people: thank you.
Contents
1
2
The Potential of Responsive Electricity Demand
15
1.1
. . . . . . . . . . . . . . . . . . . .
17
1.1.1
Existing Analyses of Island Energy Systems . .
18
1.1.2
Existing Models of Responsive Demand
19
. . . .
Policy Background and the Status Quo in Cyprus
21
2.1
Motivations for Making Demand Responsive . . . . . . . . . . . . . . . . . .
21
2.2
Obstacles to Innovation
. . . . . . . . . . . . . . . . . . . . . . . . . . . . .
26
2.2.1
Incomplete Market for Electricity . . . . . . . . . . . . . . . . . . . .
26
2.2.2
Imperfect Market Information . . . . . . . . . . . . . . . . . . . . . .
28
2.3
Event-Based Demand Response . . . . . . . . . . . . . . . . . . . . . . . . .
31
2.4
Other Approaches to Enable Responsive Demand . . . . . . . . . . . . . . .
33
2.4.1
Dynamic Electricity Pricing . . . . . . . . . . . . . . . . . . . . . . .
33
2.4.2
Energy Efficiency . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
34
2.4.3
Automated Energy Management
. . . . . . . . . . . . . . . . . . . .
36
. . . . . . . . . . . . . . . . . . . . . . . . .
37
2.5
3
Literature Review
Viability of Policy Approaches
Modeling Framework
41
3.1
WILMAR Model Overview
. . . . . . . . . . . . . . . . .
41
3.2
Approach to Modeling Responsive Loads . . . . . . . . . .
42
3.2.1
Load Parameter Definitions . . . . . . . . . . . . .
45
3.2.2
Load Shift Equations . . . . . . . . . . . . . . . . .
46
Approach to Modeling CSP with Thermal Energy Storage
50
3.3
5
4
5
. . . . . . . . . . . . . . . . . . . . . . . . . . . . .
53
. . . . . . . . . . . . . . . . . .
57
4.3
CSP Generation Parameter Values . . . . . . . . . . . . . . . . . . . . . . .
59
4.4
Solar Resource Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
60
4.5
Wind Resource Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
62
4.6
Electricity Demand Data
. . . . . . . . . . . . . . . . . . . . . . . . . . . .
63
4.1
Load Parameter Values
4.2
Conventional Generation Parameter Values
65
Results from Modeled Scenarios
5.1
Overview of Scenario Design . . . . . . . . . . . . . . . . . . . . . . . . . . .
65
5.2
Results for 2020 with Natural Gas Generation and Gradual Demand Growth
68
5.2.1
Unit Commitment Scheduling . . . . . . . . . . . . . . . . . . . . . .
68
5.2.2
Operational Patterns of Demand Response
. . . . . . . . . . . . . .
76
5.2.3
Demand Response Value Assessment . . . . . . . . . . . . . . . . . .
82
5.2.4
Impact of Fast Economic Growth . . . . . . . . . . . . . . . . . . . .
85
5.2.5
Sensitivity to Demand Response Price and Participation . . . . . . .
86
Results for 2020 with Fuel Oil Generation and Gradual Demand Growth . .
90
5.3.1
Unit Commitment Scheduling . . . . . . . . . . . . . . . . . . . . . .
90
5.3.2
Operational Patterns of Demand Response
. . . . . . . . . . . . . .
95
5.3.3
Demand Response Value Assessment . . . . . . . . . . . . . . . . . .
101
5.3.4
Impact of Fast Economic Growth . . . . . . . . . . . . . . . . . . . .
103
5.3.5
Sensitivity to Demand Response Price and Participation . . . . . . .
105
5.3
6
53
Input Data
109
Policy Implications
6.1
Limitations of Modeling Results . . . . . . . . . . . . . . . . . . . . . . . . .
109
6.2
Program Design Considerations . . . . . . . . . . . . . . . . . . . . . . . . .
111
A Abbreviations
115
B Solar Data Generation Method
117
C Additional Figures
119
.
119
C.2 Figures for 2020 with Fuel Oil Generation and Fast Demand Growth . . . .
131
C.1
Figures for 2020 with Natural Gas Generation and Fast Demand Growth
6
List of Figures
. . . . .
2-1
Projected Load Duration Curves (2020, Gradual Demand Growth)
2-2
Projected and Realized Wind Generation in Cyprus on November 18, 2012,
Adapted from an Electricity Authority of Cyprus Presentation [79] . . . . .
2-3
23
24
Projected Demand Per Hour of the Year, Before (L) and After (R) Renewables (2020, Gradual Demand Growth) . . . . . . . . . . . . . . . . . . . . .
29
3-1
A Sample Dishwasher Load Shift . . . . . . . . . . . . . . . . . . . . . . . .
43
3-2
A Sample HVAC Load Shift . . . . . . . . . . . . . . . . . . . . . . . . . . .
44
3-3
CSP+TES Modeling Approach . . . . . . . . . . . . . . . . . . . . . . . . .
50
5-1
Load Duration Curve (2020, Natural Gas, Gradual Demand Growth, No
Dem and Response) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
5-2
Variable Costs Per Unit (2020, Natural Gas, Gradual Demand Growth, DR
Price=50%, DR Participation=50%)
5-3
. . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . .
. . . . . . . . . . .
72
74
Unit Commitment with Demand Response, 7/18-7/24 (2020, Natural Gas,
Gradual Demand Growth, DR Price=50%, DR Participation=50%)
5-7
. . . .
Unit Commitment in Baseline Case, 7/18-7/24 (2020, Natural Gas, Gradual
Demand Growth, DR Price=oo, DR Participation=0%)
5-6
71
Unit Commitment with Demand Response, 1/24-1/30 (2020, Natural Gas,
Gradual Demand Growth, DR Price=50%, DR Participation=50%)
5-5
70
Unit Commitment in Baseline Case, 1/24-1/30 (2020, Natural Gas, Gradual
Demand Growth, DR Price=oc, DR Participation=0%)
5-4
69
. . . .
75
Impact of Curtailment on Demand by Hour (2020, Natural Gas, Gradual
Demand Growth, DR Price=50%, DR Participation=50%)
7
. . . . . . . . .
76
5-8
Impact of Curtailment on Load Duration Curve (2020, Natural Gas, Gradual
Demand Growth, DR Price=50%, DR Participation=50%)
5-9
. . . . . . . . .
77
Impact of Load Shifts on Load Duration Curve (2020, Natural Gas, Gradual
Demand Growth, DR Price=z50%, DR Participation=50%)
. . . . . . . . .
78
5-10 Impact of Load Shifts on Load Demand by Hour (2020, Natural Gas, Gradual
Demand Growth, DR Price=50%, DR Participation=50%)
. . . . . . . . .
79
5-11 Cumulative Impact of Load Shifts Per Start Time of Load Reduction (2020,
Natural Gas, Gradual Demand Growth, DR Price=50%, DR Participation=50%) 79
5-12 Aggregated Cumulative Operating Mode of Load Shifts (2020, Natural Gas,
Gradual Demand Growth, DR Price=50%, DR Participation=50%)
. . . .
80
5-13 Cumulative Operating Mode of Active Load Shift Programs (2020, Natural
.
Gas, Gradual Demand Growth, DR Price=50%, DR Participation=50%)
81
5-14 Impact of Demand Response on Load Factors of Dispatched Units (2020, Natural Gas, Gradual Demand Growth, DR Price=50%, DR Participation=50%)
83
5-15 Load Duration Curve (2020, Natural Gas, Fast Demand Growth, No Demand
R esponse) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
85
5-16 Normalized Average Daily Demand Response Usage (2020, Natural Gas,
Gradual Demand Growth, Stochastic Simulations)
. . . . . . . . . . . . . .
87
5-17 Normalized Increase in Efficiency Score, Weighted by Generator Capacity
(2020, Natural Gas, Gradual Demand Growth, Stochastic Simulations) .
.
.
88
5-18 Normalized CO 2 Emissions Reduction (2020, Natural Gas, Gradual Demand
Growth, Stochastic Simulations)
. . . . . . . . . . . . . . . . . . . . . . . .
88
5-19 Load Duration Curve (2020, Fuel Oil, Gradual Demand Growth, No Demand
R esponse) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
90
5-20 Variable Costs Per Unit (2020, Fuel Oil, Gradual Demand Growth, DR
. . . . . . . . . . . . . . . . . . . . . .
Price=50%, DR Participation=50%)
91
5-21 Unit Commitment in Baseline Case, 1/24-1/30 (2020, Fuel Oil, Gradual Demand Growth, DR Price=oo, DR Participation=0%) . . . . . . . . . . . . .
92
5-22 Unit Commitment with Demand Response, 1/24-1/30 (2020, Fuel Oil, Gradual Demand Growth, DR Price=50%, DR Participation=50%)
. . . . . . .
93
5-23 Unit Commitment in Baseline Case, 7/18-7/24 (2020, Fuel Oil, Gradual Demand Growth, DR Price=oo, DR Participation=0%) . . . . . . . . . . . . .
8
94
5-24 Unit Commitment with Demand Response, 7/18-7/24 (2020, Fuel Oil, Gradual Demand Growth, DR Price=50%, DR Participation=50%)
. . . . . . .
95
5-25 Impact of Curtailment on Load Duration Curve (2020, Fuel Oil, Gradual
Demand Growth, DR Price=50%, DR Participation=50%)
. . . . . . . . .
96
5-26 Impact of Load Shifts on Load Duration Curve (2020, Fuel Oil, Gradual
Demand Growth, DR Price=50%, DR Participation=50%)
. . . . . . . . .
96
5-27 Impact of Curtailment on Demand by Hour (2020, Fuel Oil, Gradual Demand
Growth, DR Price=50%, DR Participation=50%) . . . . . . . . . . . . . . .
97
5-28 Impact of Load Shifts on Load Demand by Hour (2020, Fuel Oil, Gradual
Demand Growth, DR Price=50%, DR Participation=50%)
. . . . . . . . .
98
5-29 Cumulative Impact of Load Shifts Per Start Time of Load Reduction (2020,
Fuel Oil, Gradual Demand Growth, DR Price=50%, DR Participation=50%)
99
5-30 Aggregated Cumulative Operating Mode of Load Shifts (2020, Fuel Oil,
Gradual Demand Growth, DR Price=50%, DR Participation=50%)
. . . .
100
5-31 Cumulative Operating Mode of Active Load Shift Programs (2020, Fuel Oil,
Gradual Demand Growth, DR Price=50%, DR Participation=50%)
. . . .
100
5-32 Impact of Demand Response on Load Factors of Dispatched Units (2020,
Fuel Oil, Gradual Demand Growth, DR Price=50%, DR Participation=50%) 102
5-33 Load Duration Curve (2020, Fuel Oil, Fast Demand Growth, No Demand
R esponse) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
104
5-34 Normalized Average Daily Demand Response Usage (2020, Fuel Oil, Gradual
Demand Growth, Stochastic Simulations)
. . . . . . . . . . . . . . . . . . .
105
5-35 Normalized Increase in Efficiency Score (2020, Fuel Oil, Gradual Demand
Growth, Stochastic Simulations)
. . . . . . . . . . . . . . . . . . . . . . . .
106
5-36 Normalized CO 2 Emissions Reduction (2020, Fuel Oil, Gradual Demand
Growth, Stochastic Simulations)
. . . . . . . . . . . . . . . . . . . . . . . .
107
C-1 Variable Costs Per Unit (2020, Natural Gas, Fast Demand Growth, DR
Price=50%, DR Participation=50%)
. . . . . . . . . . . . . . . . . . . . . .
120
C-2 Unit Commitment in Baseline Case, 1/24-1/30 (2020, Natural Gas, Fast
Demand Growth, DR Price=oo, DR Participation=0%)
9
. . . . . . . . . . .
121
C-3
Unit Commitment with Demand Response, 1/24-1/30 (2020, Natural Gas,
Fast Demand Growth, DR Price=50%, DR Participation=50%) . . . . . . .
C-4
122
Unit Commitment in Baseline Case, 7/18-7/24 (2020, Natural Gas, Fast
Demand Growth, DR Price=oo, DR Participation=0%)
. . . . . . . . . . .
123
C-5 Unit Commitment with Demand Response, 7/18-7/24 (2020, Natural Gas,
Fast Demand Growth, DR Price=50%, DR Participation=50%) . . . . . . .
124
C-6 Impact of Curtailment on Demand by Hour (2020, Natural Gas, Fast Demand
Growth, DR Price=50%, DR Participation=50%) . . . . . . . . . . . . . . .
125
C-7 Impact of Curtailment on Load Duration Curve (2020, Natural Gas, Fast
Demand Growth, DR Price=50%, DR Participation= 50%)
. . . . . . . . .
125
C-8 Impact of Load Shifts on Load Duration Curve (2020, Natural Gas, Fast
Demand Growth, DR Price=50%, DR Participation=50%)
. . . . . . . . .
126
C-9 Impact of Load Shifts on Load Demand by Hour (2020, Natural Gas, Fast
Demand Growth, DR Price=50%, DR Participation=50%)
. . . . . . . . .
126
C-10 Cumulative Impact of Load Shifts Per Start Time of Load Reduction (2020,
Natural Gas, Fast Demand Growth, DR Price=50%, DR Participation=50%) 127
C-11 Aggregated Cumulative Operating Mode of Load Shifts (2020, Natural Gas,
Fast Demand Growth, DR Price=50%, DR Participation=50%) . . . . . . .
127
C-12 Cumulative Operating Mode of Active Load Shift Programs (2020, Natural
Gas, Fast Demand Growth, DR Price=50%, DR Participation=50%)
. . .
128
C-13 Impact of Demand Response on Load Factors of Dispatched Units (2020,
Natural Gas, Fast Demand Growth, DR Price=50%, DR Participation=50%) 129
C-14 Normalized Average Daily Demand Response Usage (2020, Natural Gas, Fast
Demand Growth, Deterministic Simulations)
. . . . . . . . . . . . . . . . .
130
C-15 Normalized Increase in Efficiency Score (2020, Natural Gas, Fast Demand
Growth, Deterministic Simulations) . . . . . . . . . . . . . . . . . . . . . . .
130
C-16 Normalized CO 2 Emissions Reduction (2020, Natural Gas, Gradual Demand
Growth, Deterministic Simulations) . . . . . . . . . . . . . . . . . . . . . . .
130
C-17 Variable Costs Per Unit (2020, Fuel Oil, Fast Demand Growth, DR Price=50%,
DR Participation=50%)
. . . . . . . . . . . . . . . . . . . . . . . . . . . . .
131
C-18 Unit Commitment in Baseline Case, 1/24-1/30 (2020, Fuel Oil, Fast Demand
Growth, DR Price=oo, DR Participation=0%)
10
. . . . . . . . . . . . . . . .
132
C-19 Unit Commitment with Demand Response, 1/24-1/30 (2020, Fuel Oil, Fast
Demand Growth, DR Price=50%, DR Participation=50%)
. . . . . . . . .
133
C-20 Unit Commitment in Baseline Case, 7/18-7/24 (2020, Fuel Oil, Fast Demand
Growth, DR Price=oo, DR Participation=0%)
. . . . . . . . . . . . . . . .
134
C-21 Unit Commitment with Demand Response, 7/18-7/24 (2020, Fuel Oil, Fast
Demand Growth, DR Price=50%, DR Participation=50%)
. . . . . . . . .
135
C-22 Impact of Curtailment on Demand by Hour (2020, Fuel Oil, Fast Demand
Growth, DR Price=50%, DR Participation=50%) . . . . . . . . . . . . . . .
136
C-23 Impact of Curtailment on Load Duration Curve (2020, Fuel Oil, Fast Demand
Growth, DR Price=50%, DR Participation= 50%) . . . . . . . . . . . . . . .
136
C-24 Impact of Load Shifts on Load Duration Curve (2020, Fuel Oil, Fast Demand
Growth, DR Price=50%, DR Participation=50%) . . . . . . . . . . . . . . .
137
C-25 Impact of Load Shifts on Load Demand by Hour (2020, Fuel Oil, Fast Demand
Growth, DR Price=50%, DR Participation=50%) . . . . . . . . . . . . . . .
137
C-26 Cumulative Impact of Load Shifts Per Start Time of Load Reduction (2020,
Fuel Oil, Fast Demand Growth, DR Price=50%, DR Participation=50%)
.
138
. . . . . . . . .
138
C-27 Aggregated Cumulative Operating Mode of Load Shifts (2020, Fuel Oil, Fast
Demand Growth, DR Price=50%, DR Participation=50%)
C-28 Cumulative Operating Mode of Active Load Shift Programs (2020, Fuel Oil,
Fast Demand Growth, DR Price=50%, DR Participation=50%) . . . . . . .
139
C-29 Impact of Demand Response on Load Factors of Dispatched Units (2020,
Fuel Oil, Fast Demand Growth, DR Price=50%, DR Participation=50%)
.
140
C-30 Normalized Average Daily Demand Response Usage (2020, Fuel Oil, Fast
Demand Growth, Deterministic Simulations)
. . . . . . . . . . . . . . . . .
141
C-31 Normalized Increase in Efficiency Score (2020, Fuel Oil, Fast Demand Growth,
Deterministic Simulations) . . . . . . . . . . . . . . . . . . . . . . . . . . . .
141
C-32 Normalized CO 2 Emissions Reduction (2020, Fuel Oil, Fast Demand Growth,
Deterministic Simulations) . . . . . . . . . . . . . . . . . . . . . . . . . . . .
11
142
12
List of Tables
2.1
Conventional Generation Mix in 2013 and 2020 . . . . . . . . . . . . . . . .
23
2.2
Renewable Generation Mix in 2013 and 2020
. . . . . . . . . . . . . . . . .
24
3.1
Summary of Notation
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
47
4.1
Estimated Theoretical Maximum Responsive Load at 2012 Demand Levels.
54
4.2
Modeled Operational Constraints of Responsive Loads (All Units in Hours)
55
4.3
Modeled Cost Components of Responsive Load Shifts or Curtailments
. . .
56
4.4
Overview of Modeled Conventional Generators
. . . . . . . . . . . . . . . .
57
4.5
Modeled Operational Constraints of Conventional Generators . . . . . . . .
57
4.6
Modeled Variable Cost Components of Conventional Generators
. . . . . .
58
4.7
Modeled Fuel Prices . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
59
4.8
Overview of Modeled CSP Generators
. . . . . . . . . . . . . . . . . . . . .
60
4.9
Modeled Operational Constraints of CSP Generators . . . . . . . . . . . . .
60
4.10 Modeled Variable Cost Components of CSP Generators
. . . . . . . . . . .
60
4.11 Input Data Statistics for PV. Units are in MW. . . . . . . . . . . . . . . . .
61
4.12 Input Data Statistics for CSP Without Storage. Units are in MW.
. . . . .
61
. . . . . . . . . . . . . . . .
62
4.14 Demand Input Data Statistics (2012). Units are in MW. . . . . . . . . . . .
63
5.1
Stochastic (S) and Deterministic (D) Cases 1 . . . . . . . . . . . . . . . . .
67
5.2
Summary of Demand Response Impacts (2020, Natural Gas, Gradual De-
4.13 Wind Input Data Statistics. Units are in MW.
mand Growth, DR Price=50%, DR Participation=50%)
5.3
. . . . . . . . . . .
84
Summary of Demand Response Impacts (2020, Natural Gas, Fast Demand
Growth, DR Price=50%, DR Participation=50%) . . . . . . . . . . . . . . .
13
86
5.4
Summary of Demand Response Impacts (2020, Fuel Oil, Gradual Demand
Growth, DR Price=50%, DR Participation=50%) . . . . . . . . . . . . . . .
5.5
A.1
103
Summary of Demand Response Impacts (2020, Fuel Oil, Fast Demand Growth,
DR Price=50%, DR Participation=50%) . . . . . . . . . . . . . . . . . . . .
104
Abbreviations Used in the Text . . . . . . . . . . . . . . . . . . . . . . . . .
115
14
Chapter 1
The Potential of Responsive
Electricity Demand
The electricity sector of the Mediterranean island nation of Cyprus will undergo two major
planned transformations in the near future.
One transformation has been sparked by a
European Union mandate for Cyprus to drastically increase its usage of renewable energy
from roughly 5% at present to 16% by the year 2020.[74] The shift to greater penetration
of renewables is likely to lower electricity costs and lessen the environmental impact of consumption in Cyprus, but it is also likely to stress the power system by introducing large
amounts of intermittent and unpredictable electricity. Another transformation may result
from natural gas deposits recently found in Cypriot undersea territory. Like the usage of
renewables, a shift from today's reliance on imported fuel oil to the planned usage of domestic natural gas is also likely to lower electricity costs, but it will change the configuration
of existing power plants in Cyprus, and it will also change their operational behavior when
dispatched to minimize costs.
These opportunities and challenges in the energy sector are further complicated by a
recent financial and economic crisis in Cyprus. The crisis may require a lengthy, gradual
recovery, and it will certainly increase the value of low-cost solutions in the electricity sector,
where Cypriot households already pay one of the highest average electricity prices in the
EU. 1 [31]
'Throughout this thesis, "Cyprus" is used to refer specifically to the Republic of Cyprus, not including
the northern side of a de facto partition of the island of Cyprus. Only the electricity supply and demand of
the southern side of the partition are modeled here.
15
This work assesses the impacts of making electricity demand more responsive to the
condition of the Cypriot electricity system. By making demand responsive in systems with
high levels of renewable energy-that is, by creating ways to modify electricity consumption
patterns to reduce stresses on the electricity system-it may be possible to further lower
the cost of electricity generation, reduce fuel usage and carbon emissions, and increase
the usable output from installed renewables.
This work focuses on event-based demand
response programs, which are enrollment-based programs which incentivize energy users
to modify their consumption patterns on occasions when they receive a request from the
program operator. The program could be created and managed by the Electricity Authority
of Cyprus (EAC), which is the nation's sole electric utility; the grid operator; or a private
company.
In order to assess the operational impacts of responsive demand in Cyprus, this work
constructs an optimization model of the Cypriot generation system and the scheduling of
each generator to produce a given output at a given hour, known as the unit commitment
schedule. The model includes generation from conventional sources, wind power, photovoltaic (PV) solar power, and concentrating solar power (CSP); some CSP generation is
equipped with thermal energy storage, which makes it more controllable and flexible. The
model uses a linear programming methodology, including parameters and constraints specific
to each modeled electricity end use. End uses include loads in the residential, commercial,
and desalination sectors, some of which can be fully curtailed and some of which can only
be delayed or advanced in time. The model uses historical demand and wind data, scaled
to match 2020 projections. Dr. Constantinos Varnavas at the Generation Unit of the EAC
provides certain operational constraints for installed conventional generators. Dr. Alexandros G. Charalambides at the Department of Environmental Science and Technology of the
Cyprus University of Technology provides data regarding solar generation potential.
The model is used to assess impacts of responsive demand in 2020 both with and without
a transition to natural gas generation, with either a gradual or fast economic recovery
between now and then. In each of the modeled scenarios, a demand response program
including 50% of potential participants which pays participants at a rate of 50% the cost of
the system's most expensive generator is found to:
9 Reduce average generation costs by 1.1%-2.1% after accounting for payments to demand response program participants, which make up less than one-tenth of one percent
16
of total costs
" Completely eliminate the curtailment of usable wind generation
" Increase the usable output of CSP plants by 0.7%-1.3%
" Reduce fuel usage by 1.1%-1.4%
" Reduce carbon dioxide emissions by 1.1%-1.5%
This chapter introduces this thesis within the context of the existing literature. Chapter
2 outlines the current context of Cyprus and the motivations for making demand responsive, highlights several market obstacles to demand-side innovations, and describes several
approaches to responsive demand, including event-based demand response programs. Chapter 3 describes the methodology used to model the Cyprus power system with responsive
demand, and Chapter 4 describes the input data used in the model. Chapter 5 presents the
results from each modeled scenario of Cyprus in 2020, and Chapter 6 synthesizes the results
across all the scenarios into several general policy insights for demand response program
design in Cyprus.
1.1
Literature Review
Prior work exists regarding responsive electricity demand's potential to help meet islands'
power needs, in Cyprus and elsewhere. One operational model of responsive demand on
islands, which this work builds upon, is described by Critz, Connors, and Busch. [23] It
assesses the potential of event-based demand response in Hawaii, and finds that it could
reduce power system costs, reduce the curtailment of wind power, and eliminate reserve
deficits.
One interesting insight was that the option value of demand response can be
greater than the operational usage might imply; the results shown above required only 18
calls to demand response over one year.
Kritz et al use the Wind Integration in Liberalized Markets (WILMAR) model, developed by Peter Meibom et al at Denmark's Riso National Lab, to model the island system's
unit commitment scheme with and without demand response. [62] It applies WILMAR to
Hawaii's power system to assess demand response's to balance high penetration of wind. It
divides demand response resources evenly into two varieties. One represents automatically
17
dispatchable fast-response loads that may be present in residences, while the other represents loads requiring human intervention; the latter requires a longer lead time to schedule,
but the former is assigned a 25% cost premium. It models demand response by leveraging
WILMAR's built-in capability to model storage devices. This treats demand response as a
fully shiftable resource subject to demand response program capacity constraints and operational costs. This study differs from [23] in that a set of parameters and constraints are
included to reflect the operational constraints of specific end uses more closely.
These are a few studies related to responsive demand and grid balancing on islands with
renewable generation, but there are a wide range of studies on each component of that topic
(island energy systems modeling and responsive demand modeling).
1.1.1
Existing Analyses of Island Energy Systems
A body of literature has formed around the unique challenges faced in island power systems.
Many of these challenges, such as grid stability, high fuel costs, and goals to drastically
increase renewable usage, are applicable to Cyprus and this work.
Studies on grid balancing and and stability specific to Cyprus exist.
A 1995 Elec-
tricity Authority of Cyprus (EAC) study assesses central coordination of under-frequency
load shedding (UFLS) as a last-resort measure for protection against sudden power system
disruptions on islands such as Cyprus, and recommends that load shedding programs including 80% of Cyprus's total electricity load be developed. [19] More recently, Andrianesis,
Liberopoulos, and Varnavas [57] assess a range of impacts higher levels of wind generation
may have on the Cyprus power system, including the requirements for operating reserves
for generation and overall operational costs. Similarly, De Vos et al [80] analyze Cyprus
with high levels of wind penetration and find that the island's reserve requirements should
be increased. Hansen and Papalexopoulos [35] perform a similar analysis of the island of
Crete, Cyprus's neighbor in the Mediterranean. All three of these recent studies find that
with sufficiently high levels of wind penetration, significant amounts of wind generation will
need to be curtailed unless some form of energy storage is implemented; De Vos et al also
predict the necessity of shedding loads to maintain grid stability.
Several recent studies of the Azores islands also addresses other challenges shared by
many island power systems. Livengood et al performed a high-level analysis of responsive
demand in isolated grids.
[24] The analysis focused on a case study of using responsive
18
demand to displace diesel generation in the island of Flores in the Azores. The study assessed
diesel oil displacement that could be achieved using a high-level mathematical comparison
of each option's power capacity. The study found making 5% of demand responsive could
displace as much diesel as if a hydroelectric plant on the island were doubled in size.
Parness uses a unit commitment model to assess impacts of vehicle electrification in the
Azores. [59] Parness finds that "optimal" charging of electric vehicles, perhaps via central
coordination, significantly reduces power system operational costs, fuel usage, and carbon
emissions. In some regards, the impacts of electric vehicles are similar to the impacts of
responsive demand as described in this work; electric vehicles are a substantial electricity
load, and Parness's work illustrates how coordination of demand benefits islands.
DeAmicis performs a unit commitment study of the use of electricity storage to balance
wind power in the Azores, finding that storage could increase renewable generation by 10%
and reduce production costs by 16%. [25] In some regards, storage and responsive demand
are substitutes in power systems; each can balance intermittent generation by lowering peak
demand, "filling in" troughs in demand, and reacting quickly to fluctuations.
1.1.2
Existing Models of Responsive Demand
Jason Black's 2005 dissertation is a comprehensive assessment of the technical, economic,
and regulatory considerations associated with responsive loads.
[8] The work includes a
system dynamics model of responsive electricity loads, with long-term technology improvements, learning by consumers, policy changes, and changes in demand elasticity. Many
of its findings, including long-term distributional and environmental effects of demand response, are out of this work's scope but warrant further analysis in the Cyprus context and
elsewhere.
One category of responsive demand model exists at the building scale. Livengood and,
more recently, Leow have used dynamic programming to build operational models of responsive demand for households. Livengood's work models an "Energy Box" which can locally
coordinate a building's appliance-based power consumption, distributed generation, inhabitant preferences, and responses to real-time price signals. [41] Leow's work further develops
the model to include zonal heating and air conditioning of buildings, automatic learning of
occupancy patterns, and manual occupancy updates via smart phones. [40] Livengood and
Leow's works fall into the category of household-level responsive demand modeling. Stadler
19
examines the usefulness of modifying usage of thermal loads within buildings, and finds that
thermal loads can significantly ease integration of renewables. [72]
Grid-level unit commitment modeling forms a second category of responsive demand
modeling, the literature of which is vast. Numerous modeling approaches exist for unit
commitment, including linear and integer optimization, dynamic programming, system dynamics, neural networks, and agent-based modeling. [12, 45, 69, 65] Heuristic and evolutionary approaches have also been applied to unit commitment, ranging from simulated
annealing to methods drawn from the behavior of bacteria. [84, 30, 76, 68]
Compared to the wider body of unit commitment studies, those which include responsive
demand are much less numerous. Many focus on event-based demand response to balance
renewables, including Critz et al. [23] Zhao and Zeng use the cutting plane algorithm to
model demand response as a time-based demand elasticity which responds to price signals.
[83] They find that greenhouse gas emissions are reduced and power generation is operated
at lower cost.
This work models unit commitment to balance supply and demand at the grid scale,
but models specific types of end uses at a level of detail similar to that of its conventional
generation units. The model lacks the distributed decision-making often reflected in agentbased models and some building-scale models, making it most suitable for modeling centrally
coordinated event-based demand response, described further in Section 2.3. This work aims
to further develop the literature on responsive demand in systems with significant renewable
penetration, and to create insights useful to Cyprus and other isolated power systems.
20
Chapter 2
Policy Background and the Status
Quo in Cyprus
2.1
Motivations for Making Demand Responsive
Power systems must constantly balance electricity supply and demand in order to meet consumer expectations and prevent grid instabilities. Responsive demand has the potential to
assist that balancing, particularly in light of recent security of supply issues with Cyprus's
electricity generation system and upcoming increases in supply intermittency. Like most
island power systems, Cyprus must generate all electricity locally, which reduces energy
supply security and stability. Per [74], 95% of this electricity currently comes from burning
heavy fuel oil, which is imported from international sources at high cost. One source of
supply-side insecurity comes from the availability and price of oil, most of which is purchased from Syria and the Russian Federation according to [56]. The historical impact of
this insecurity has been largely limited to rate fluctuations, as power bills are adjusted each
month to account for the current price of fuel. However, the impacts of two recent developments on the supply side-a catastrophe near the Vasilikos power plant at the Evangelos
Florakis Naval Base, and growing solar and wind generation-are more drastic.
On July 11, 2011, a large stockpile of explosive munitions stored at the Evangelos Florakis Naval Base self-detonated suddenly and unexpectedly. The explosion killed 13 people,
injured 62, and damaged many houses in a nearby village. The nearby Vasilikos Power
Station was severely damaged, thus taking roughly half of Cyprus's generation capacity of21
fline. For the following month, widespread rolling residential blackouts were necessary due
to supply shortages. For a longer period of time, Cyprus also had to lease very expensive
temporary generators from Israel. This cost, coupled with nearly one billion dollars of repair
costs for the Vasilikos plant, drastically raised the cost of electricity supply, and [34] notes
the total economic impact of the catastrophe has been estimated at 2.83 billion US dollars.
More broadly, the Vasilikos catastrophe demonstrated that no form of electricity generation is truly fully reliable and controllable. That is why power systems are typically built to
meet the peak level of aggregate demand required at any instant, plus some safety margin
to accommodate generator outages and demand increases. Because it is an island, Cyprus's
safety margin is even more relevant; there is no larger power grid connected to the Cyprus
grid that can provide backup generation.
One way to increase a power system's safety margin is to lower consumption at hours
of peak demand by shifting it forward or backward in time; lowering peak consumption is
one potential benefit to making demand more responsive. Figure 2-1 shows the projected
load duration curves of Cyprus in 2020, the number of hours expected to have a given level
of demand before and after projected levels of renewable generation are taken into account.
The curves were made by scaling historical data from 2012 to match projected 2020 demand
levels, assuming gradual demand growth.'.
The highest level of demand in a given year in Cyprus occurs for a very brief amount of
time; this is because demand increases exponentially when tourism and high temperatures
coincide. In fact, if only 1.58 gigawatt-hours, or 0.04%, of total energy demand are shifted
in time or curtailed to lower peak demand, the projected 2020 peak demand to be met
after factoring in renewable generation could be 10% lower. This would allow more of the
installed conventional generation capacity to serve as a safety margin, rather than being used
for scheduled generation. In a year such as 2011 in Cyprus, in which available generation
was lower than expected, the scale of the rolling blackouts might have been reduced. The
value of responsive demand as a partial mitigation for low safety margins is high in Cyprus
largely because it is an island; this has been seen in other isolated power systems. [23, 21, 24]
Renewable generation is another source of unpredictability in Cyprus's energy supply. In
order to meet European Union regulations, Cyprus plans to generate at least 16% percent of
its electricity from renewable sources by 2020. The island currently generates only roughly
'Details of demand assumptions used for this work are given in Section 4.6.
22
1000
I
I
I
I
I
I
I
I
Demand
Demand minus Wind
Demand minus Wind and PV
Demand minus Wind, PV, and CSP
900
800
700
600
.0
500
400
300
200
100 0
0
1000
2000
3000
4000
5000
Hours
6000
7000
8000
Figure 2-1: Projected Load Duration Curves
(2020, Gradual Demand Growth)
5% of its energy from renewables, the majority of which comes from
an 82 megawatt (MW)
wind farm at Orites, the first phase of two at that site. As shown in
Table 2.2, by 2020,
the island plans to increase its installed wind capacity from 135 MW to 300 MW and its
installed solar capacity from 9 MW to 267 MW.
Energy Source
Heavy Fuel Oil
#2 Diesel Oil/Gas Oil
Installed Capacity,
2013 (MW)
852
628
Natural Gas
Total
Estimated Capacity,
2020 (MW)
360
290
0
1152
1480
1802
Table 2.1: Conventional Generation Mix in 2013 and 2020
Relative to the slow and steady evolution over many decades that characterize many
energy systems, these transformations are rapid and drastic. Solar and wind power are inherently intermittent, and also subject to forecasting errors. This increasing unpredictability
of supply may stress the grid's ability to balance supply and demand. Figure 2-2 shows
actual recorded wind generation data from Cyprus for one day. In some
stretches of time,
23
Installed Capacity,
2013 (MW)
Biomass
10
Estimated Capacity,
2020 (MW)
17
Wind
135
300
9
0
192
75
154
584
Energy Source
Solar Photovoltaic
Concentrating Solar Power
Total
Table 2.2: Renewable Generation Mix in 2013 and 2020
such as from 5AM to 12PM, actual wind generation follows the predicted generation relatively closely, but at other times, wind forecasting is much less accurate. At 2AM the actual
wind generation is less than half the projected value. At 12PM, just 10 hours later, wind
generation spikes up to four times the predicted level for less than an hour before falling
then spiking again.
120
100
80
60
40 -
20 -
0
12AM
4AM
8AM
12PM
-4- Predicted Wind (MW)
4PM
-
8PM
12AM
Realized Wind (MW)
Figure 2-2: Projected and Realized Wind Generation in Cyprus on November 18, 2012,
Adapted from an Electricity Authority of Cyprus Presentation [79]
The typical method of mitigating this intermittency and unpredictability is to run a set
of conventional thermal generators at a low standby level-the most inefficient output levelso they can react quickly to fluctuations.
Open cycle gas turbine "peaking" generators,
so named because their low capital costs (and relatively low efficiency) are designed to
occasionally supply peak load rather than constantly supplying base load around the clock,
sometimes are used in this fashion due to their fast ramping rates from low to high output
and vice versa. In some contexts, including Cyprus, other units, including those that would
typically supply base load, are run at a lower output level so they can be ramped up to a
higher level and balance sudden increases in demand or decreases in intermittent generation.
24
In some cases, as assessed by [25] to cite one example, intermittency could be mitigated somewhat by the use of storage in the form of pumped water, batteries or flywheels
to smooth power supply fluctuations. Unfortunately, it is not clear that Cyprus's water
resources are sufficient for large-scale pumped water storage. [11] finds that severe water
shortages are common and daily water consumption per capita is one fourth that of some
other developed countries. Furthermore, energy storage of any form at the bulk power scale
is often relatively expensive; in [50], the Cyprus TSO suggests that current capital costs
may be too high for it to be implemented without new incentives of some kind. Cyprus
electricity prices are already the highest in the EU in terms of purchasing power standard
(PPS), per [52], and cost will be a key concern in Cyprus's approach to managing supply
unpredictability and intermittency.
Another option to increase system safty margins and help balance intermittent generation could be to connect Cyprus to a larger system via an undersea power cable, which
could provide backup generation and perhaps balancing services. An idea for such a project,
labeled the Euro-Asia Interconnector, is currently being considered by a coalition of stakeholders in Cyprus, Greece, and Israel; the future prices of electricity in each nation would
determine which countries become net exporters or importers on the interconnect.
The
results of [33], a recent feasibility study were promising, and the three nations are currently
considering whether and how to fund the project. However, the undersea cable would be
the deepest ever built, at 2700 meters at one point, and [39] notes that cost estimates have
increased from 4 1.5 billion to 6 2.8 billion; it remains to be seen whether the recovering
economies of Cyprus and Greece can handle those costs. 2
Making demand more responsive-that is, enabling energy use to be shifted, curtailed
or, in some cases, increased to balance fluctuating wind and solar-is another option to
reduce the need to operate inefficient conventional generation at inefficient output levels.
That could reduce power costs and lower carbon emissions (another EU mandate, as noted
in [66]).
Compared to the use of centralized storage or a mainland grid interconnect to
improve supply security and balance intermittency, it is likely that the capital costs to enable
large-scale responsive demand are less expensive. Furthermore, the sources of responsive
2
1f the interconnect is built, issues of supply security would be reduced but not eliminated, because
of the
chance of the interconnect itself having periodic outages. Issues of balancing intermittency may still remain,
depending on the economics of using long-distance transmission for grid balancing; it may be the case that
responsive demand or locally-installed storage would remain a less expensive option.
25
demand-i.e., energy consumers across the nation-may be more decentralized, which could
reduce inefficiencies from transmission system congestion and transmission losses over long
distances. Recent studies of the power systems of Denmark, Hawaii, and the Azores have
each shown that making demand responsive can be a relatively economical option for grid
stability and balancing. [23, 21, 24]
Some degree of supply unreliability and intermittency is unavoidable. Every generator
in Cyprus has a non-zero chance of failure at any given time, and the availability of wind
and sunlight will inevitably fluctuate. Furthermore, [10] notes that small isolated systems
like Cyprus are inherently more "vulnerable to serious disturbances" due to "lower inertia,
limited reserves, and lack of access to off-system assistance", which reaffirms the need to
make demand responsive to the grid's needs.
2.2
Obstacles to Innovation
The potential of demand-side innovations in Cyprus has not yet been realized. By default,
consumers pay a volumetric electricity bill calculated based on total electricity consumed
that month, with no consideration of when or how the consumption occurs. This means
there is little incentive for consumers to shift consumption to lower system costs or improve
grid stability. There is also not much infrastructure in place to inform consumers how much
power they tend to use at given times of the day or year or how their consumption patterns
could change to reduce system costs; the main feedback they receive is the price signal
from their monthly bill. These shortcomings draw attention to structural issues of Cyprus's
electricity market, shared with electricity markets in other systems, which could become an
impediment to demand-side innovations.
2.2.1
Incomplete Market for Electricity
The concentrated nature of Cyprus's electricity generation may be one obstacle to enabling
more responsive demand, although the evidence available suggests this obstacle may be
secondary. This market concentration is not universal in power systems around the world,
but it is common in island power systems. Massive economies of scale are needed to create
the capital-intensive infrastructure to supply electricity, and natural monopolies often arise,
particularly in small-scale or isolated power systems such as Cyprus. In fact, this is why
26
the European Union has exempted Cyprus from certain aspects of liberalization. Still, even
though these natural monopolies exist for perfectly valid reasons, there is a risk that the
lack of competition in electricity supply their incentive to invest in demand-side innovations
that could reduce electricity costs and carbon emissions.
The concentration of electricity supply in Cyprus is clear.
Until 2004, the Electric-
ity Authority of Cyprus (EAC) was the only seller in a closed electricity market; at that
point, 35% of the market was officially opened. In 2009 the market was opened with the
exception of residential households, and plans exist to fully liberalize the market by 2014
(Cyprus obtained an exemption from the European Commission allowing it to delay full
liberalization until that date). Despite this de jure liberalization, the market is still largely
de facto concentrated. EAC completely owns the transmission and distribution networks; a
recently established independent transmission system operator (TSO) exists, but EAC operates the distribution network. In other words, electricity provision is largely vertically and
horizontally integrated. Cyprus has no functioning market for electricity in the day-ahead
or intraday time frames, and purchase agreements must be negotiated bilaterally between
EAC and the TSO. [13]
There is a current controversy in Cyprus regarding the question of whether EAC exercises
its market power in an undue fashion. Electricity is expensive in Cyprus, and frustrated
citizens quoted in [61] allege that the high rates are due to EAC's monopoly. On the other
hand, EAC notes in [48] that high rates are reflective of the high costs throughout the
energy supply chain: imported oil is extremely costly, the national government has been
slow to allow creation of a less costly natural gas supply, and a fraction of the monthly
bill is a surcharge to cover a government-mandated rate paid to independent producers of
renewable generation. Furthermore, a formula approved by the Cyprus Energy Regulatory
Authority (CERA) links electricity prices to the cost of imported oil, as noted by [82]. It is
not obvious from available data that the EAC has exercised market power unfairly; it may
simply be operating as efficiently as it can in a highly constrained position.
A more general, and perhaps less controversial, concern related to EAC's market dominance is their resulting lack of incentive to innovate in new demand-side innovations. If the
market functioned with perfect competition, EAC would undertake various economically
rational actions to reduce generation costs and maintain market share.
Studies such as
[23] have shown that modest levels of responsive demand can easily pay back the required
27
investments by allowing existing conventional thermal generation to run more efficiently,
and by allowing renewables to be used to their maximum potential. Without the threat
of competition to motivate EAC to attempt such a strategy, the utility may have little
incentive to be a leader of any such demand-side innovation; a business-as-usual strategy
could yield an equally favorable market.
One promising development in favor of competitive electricity generation lies in the
national policy for integrating distributed generation from renewables.
The independent
TSO is obligated to offer a grid connection to independent renewable generators if they
meet certain technical requirements, and a feed-in tariff ensures a return on investment
for the independent generators.
These incentives, coupled with a climate conducive to
renewable generation-including very strong potential for solar generation compared to the
rest of Europe, as noted by [11]-should help bring online distributed renewables. However,
[42] finds that bureaucratic hurdles have been a major impediment to distributed generation.
Approval is needed from 19 different offices simply to get a building permit, the first of many
steps in gaining a grid connection. There are no documented rejections of licenses by the
TSO to date, but delays are severe, which has the potential to discourage investment.
Even if competition from renewable generation does not increase the utility's incentive
to innovate, a concentrated electricity supply market is not necessarily the market failure
of most concern to developing responsive demand. More fundamental aspects of the market's structure creating imperfect information on the demand side also hinder development
of responsive demand. This problem of imperfect information is underlying cause of today's status quo in Cyprus and other electricity markets-relatively inflexible, unresponsive
demand-and it warrants further examination.
2.2.2
Imperfect Market Information
In Cyprus, as in many energy systems worldwide, the end consumer is not made aware of the
real cost to create the electricity they consume at any given time. Rather, the consumer's
monthly bill is based on the total energy usage that month, regardless of when the energy
was consumed. If the consumer uses a given amount of energy heavily during peak hoursrunning an air conditioner in the late afternoon, for example-they are charged the same
amount as if they had used that energy during the middle of the night, when very few other
people are consuming much energy. There are two tariff scheme alternatives for consumers
28
who wish to opt into coarse, two-tiered time-of-use tariffs-one in which there is a special
low rate for overnight hours, and a separate rate structure option in which there is a special
high rate for midday peak hours-but as noted above, flat pricing is the default. [49] It is not
clear from available evidence whether the lack of participation in alternate tariff schemes is
due to poor publicity on the part of EAC, consumer risk-aversion, or other factors.
This flat pricing structure is far from optimally efficient, because the real costs of energy
are not flat. During peak hours, generators are operating at full capacity (sometimes including extremely expensive "peaking" generators). At other times, transmission lines may
be congested.
At still other times, there may be a surplus or deficit of generation from
renewables which cause an excess or shortfall of available electricity supply. Aside from
these effects at a relatively short, hourly timescale, the power system must operate more or
less efficiently different days of the week, since energy is consumed differently on weekends
than on weekdays. There are also seasonal variations in power system operation due to
climate and tourism patterns, and unpredictable variations in imported fuel costs.
MW
MW
900
3/1
800
-
900
3/1
700
5/1
700
5/1
600
7/1
500
*600
7/1
500
400
9/1
300
400
9/1
300
200
11/1
200
11/1
100
12/31
00:00
06:00
-0
12:00
18:00
Time of Day
100
12/31
00:00
24:00
06:00
-0
12:00
18:00
Time of Day
24:00
Figure 2-3: Projected Demand Per Hour of the Year,
Before (L) and After (R) Renewables
(2020, Gradual Demand Growth)
These variations are shown above in Figure 2-3, a graph of forecasted daily electricity
demand profiles from January 1, 2020 to December 31, 2020, scaled based on historical data
from 2012 and projections of gradual demand growth between now and 2020. Demand is
much higher than normal in a few of the hottest weeks of summer, which coincide with peak
29
tourist season. This demand is largely driven by heavy air conditioner usage; similar, in the
coolest months of the winter, peak demand is largely driven by electric heater usage once
workers return home for the evening. These seasonal and hour-by-hour demand changes
affect the level of strain on the electric generation, transmission, and distribution systems,
thus affecting system costs. None of these effects are communicated in a flat rate price, and
many of them are not conveyed via the optional two-tier pricing schedule.
In an idealized electricity market operating in a fully efficient manner, generators and
consumers would negotiate the price of each kilowatt-hour exchanged to reflect the current
costs of supplying electricity and the current demand level. When costs of generation were
higher, the market price would reflect so, and the consumer would either consume less or
pay more money to consume the same amount. Similarly, when costs were low, market
prices would also be low (or negative in some cases, such as if the utility prefers to raise
demand rather than quickly ramp down a large generator), and the consumer would be more
inclined to consume. A flat electric rate, as used in Cyprus at present, is the antithesis
of this market structure.
The consumer simply pays a rate determined by CERA. The
relatively opaque translating of one's energy usage habits into a monthly bill helps explain
the public's frustration with EAC. The lack of such price information is a form of imperfect
information: the consumer's monthly bill should be proportional, at least in theory, to the
costs of running the Cyprus power system to generate the power they used, but there is no
signaling mechanism (price-based or otherwise) to inform them of the impact their behavior
has on the system operation costs.
The value of fixing this lack of signaling is clearer in a hypothetical scenario. If a person
were to arrive home from work at 5 P.M. and then start a load of laundry to prepare for the
next workday, they would be consuming energy at the least efficient time of the day, when
the expensive "peaking" generators must be utilized. If they were to start the laundry at
8 P.M. instead, when overall demand levels are lower, there might be no negative impact
on their personal welfare. The power grid could operate more efficiently by relying less on
"peaking" generators and more on relatively cheap "base load" generation, thus lowering
total system costs. This transaction would be a Pareto improvement; overall welfare would
be improved without reducing any individual's welfare.
However, the lack of complete
information prevents the transaction from being carried out. The result is essentially a
cross-subsidy in which the consumer who does laundry during peak hours is subsidized by
30
the consumer who does laundry power later in the evening, as noted by [9]. The negative
or positive impacts of different energy consumption patterns are pricing externalities, and
fixing the imperfect information problem on the demand side could allow for those impacts
to be internalized into the market price.
To be clear, the shielding of the consumer from the actual cost of generation has a valid
historical basis.
[32] notes that energy usage transactions are not frictionless, violating
an assumption of many traditional economic analyses. The consumer would tire quickly
of having to adjust their consumption patterns minute by minute to match the whims of
the power grid and many households are empty (or the occupants are asleep) much of
the day. Having to do so may well decrease his or her overall personal economic welfare.
It is also a technical challenge for a fair market price to be set rapidly when there is no
truly functional market, and no two-way communication between buyers and sellers; even
calculating the cost of generating the electricity at a specific instant may be hard for the
utility to do. Furthermore, exposing the end user to instantaneous price fluctuations might
be unfair, since distributional effects arise when certain consumers can react more efficiently
than others. There may be a risk that dynamic prices would encourage suppliers to change
prices faster than consumers could react, although [71] notes that flat-rate designs also
offer a "temptation for generators to manipulate the market" in other ways. In all of these
respects, flat rate pricing does appear to "insulate consumers from volatility", as noted in
[8].
There are range of automated energy management technologies under research and development, as described in Section 2.4.3. Still, as the U.S. Federal Energy Regulatory Commission has detailed in [14], without a market structure creating an incentive for demand
to respond to supply-side signals, many market price-based responsive demand technologies
are largely impotent.
2.3
Event-Based Demand Response
Rather than focusing on market price-based implementations of responsive demand-for
example, real-time pricing schemes-the model described in this work focuses on eventbased demand response programs. In such programs, end users typically receive a fixed
incentive or an electricity rate reduction for enrolling in the program. In exchange, they
31
may be summoned to curtail or shift electricity usage periodically as circumstances dictate.
End users may also receive an extra payment for each event in which they are called by
the demand response program. For certain electricity consumers, such as energy-intensive
industrial sites, direct load control infrastructure may be installed on site, to remove the
need for human communication and making response times virtually instantaneous. These
programs can be coordinated by electric utilities, independent system operators, or private
companies. The business models behind them is described in more depth in [21].
There is a recent precedent in Hawaii for using event-based demand response to help
integrate intermittent generation. The Hawaiian Electric Company, facing goals of 40%
renewable electricity generation by 2030, has created a demand response intiative described
at [15]. The initiative will include residential, commercial and industrial energy consumers,
although the initial pilot phase focuses on commercial and industrial consumers with at
least 25 kilowatts of shiftable or curtailable consumption capacity. Honeywell is a partner
in the program, assisting with the buildout of a semi-automated load control infrastructure.
The Hawaiian Electric Company lists six system benefits to its demand response program:
0 "Reduce the system peak, lowering costs for all customers
* Stabilize grid frequency
o Prevent or fix grid emergencies
o Reduce the amount of imported fossil fuel used to generate electricity
o Improve the ability for Hawaiian Electric to use renewable energy
o Keep the grid running reliably and efficiently" [18]
From a modeling perspective, focusing on these event-based programs captures the type
of responsive demand most feasible to implement in the near term. It can grow incrementally
from a small to a large participation level, it requires relatively little up-front infrastructure
buildout, it can be implemented without sweeping changes to electricity market or pricing
structures, and it can take an opt-in approach with little unwanted disruption to the status
quo for end users.
32
2.4
Other Approaches to Enable Responsive Demand
There are at least three other categories of demand-side energy policy that are also potentially useful in Cyprus: policies related to dynamic electricity pricing, automated energy
management, and energy efficiency. Initiatives in any one of the categories can have impacts, but the interplay between categories is considerable. For example, pricing schemes
affect the usefulness of automated energy management, and automated energy management technologies can act as a substitute for other efficiency technologies. Because of this,
Cypriot policymakers should coordinate strategy across all three categories. In each category, Cyprus's electricity regulator and utility have made cursory steps, but there is still
room for improvement.
2.4.1
Dynamic Electricity Pricing
The specific type of energy management technology to subsidize depends on the policies
regulating electricity markets and pricing; this is why efficiency and building management
technology policies should be coupled with changes to the pricing structures.
As noted
above, market restructuring policies to address imperfect information issues and the related
pricing externalities due to flat electricity rates are of the most importance; increasing
competition should be considered secondary. Top-down command-and-control policies to
regulate energy demand are likely to be politically infeasible. They clearly have a role-such
as when rolling blackouts were arranged following the Vasilikos catastrophe, and targeted
at residential users to minimize overall economic impact-but seem inadvisable in the dayto-day grid operations. Aside from event-based demand response, other robust policies to
make demand more responsive might focus on, for instance, dynamic pricing schemes which
essentially enable economic transfer payments between energy users at different hours of
the day. Aside from this ambitious goal, policies could at least use taxes or subsidies to
better incentivize the "good behavior" of responding to the conditions of the grid.
The most direct way of addressing the externalities arising from imperfect information
on the demand side is to expose energy consumers to prices more reflective of actual grid
conditions. This could take the form of real-time pricing (RTP), in which the independent
TSO could send pricing signals directly to households.
This could take various forms;
typically, prices are published no more often than once per hour, and price forecasts are
33
typically provided to customers in advance of the hourly updates. This is perhaps the most
efficient type of pricing scheme, since it most closely ties electricity price to generation costs
and "captures a far larger share of the variation" in supply-side costs compared to other
schemes. However, the setup costs are high compared to other options; RTP participants
must have a meter capable of receiving instantaneous pricing signals, whether via a cellular
antenna, the Internet, or some other communications method.
Ideally, such a program
participant would also have an automated energy management system responding to those
price signals and adjusting each appliance's usage in real time.
If such a program were deemed impractical due to the need for advanced metering in
houses, a less capital-intensive but also less optimal option is to institute time of use (TOU)
pricing. In this scenario, each hour of the day would have a pre-determined price regardless
of the day of year; for instance, rates at 5 PM of every day could be set at a certain level.
Under TOU pricing, pricing schedules are set based on historical average, and need only be
communicated once via mail or some other low-tech method, so no two-way communication
is needed.
There are variations on this. For example, one TOU pricing schedule with
drastically higher prices could be used for the hottest, most energy-intensive weeks of the
summer and another could be used for the rest of the year. There could also be different
weekday and weekend pricing schedules.
However, there is a basic tradeoff between the
number of different pricing schedules and the ability of the consumer to understand and
respond to them. Because costs are not simply dependent on the day of the week or the
hour of the day, this is still less optimal than RTP schemes. After all, wind and solar can
fluctuate, both on an hourly and a daily timescale; these effects on generation costs would
not be internalized and the benefits of the program would be reduced. One survey found
that if given the choice between various forms of dynamic pricing or TOU pricing, 86%
of respondents would prefer a form of TOU pricing, while only 14% would prefer dynamic
prices. [77]
2.4.2
Energy Efficiency
Energy efficiency policies such as standards for building efficiency can help the end consumer
use less energy to achieve the same level of comfort without modifying their behavior.
Arguably, they are a form of responsive demand over decades-long, building investment cycle
time frames: efficiency policies reduce stress on the grid and forward multiple power system
34
policy objectives-reduced emissions, greater stability, and potentially lower electricity billsby changing demand levels.
According to [56], Cyprus's policy for efficient energy usage consists largely of as a set
of subsidies and mandates to encourage "the introduction... of more energy efficient electric
appliances and the improvement of the energy behavior of buildings" in the residential and
service sectors.
For example, in 2009, the government's provided five free compact fluo-
rescent lamps to every household and provided subsidies to improve building insulation.
There are also several command-and-control style mandates in place, such as mandatory
installation of solar thermal water heaters in new houses (due to strong sunlight and historical reasons, one in every five existing houses has a solar thermal water heater rather
than a gas- or electricity-powered one). Similarly, periodic maintenance and inspection of
air conditioning systems is mandated to ensure the systems are performing correctly and
efficiently.
These efficiency subsidies and mandates are a necessary first step towards enabling
energy consumers to behave more optimally in the larger power system over long timescales.
A recent study showed that much of the island's energy usage surge in recent history has
been due to more widespread use of electric appliances-largely those used for heating and
cooling buildings-which are not designed for efficiency. This surge is largely driven by
increasing temperatures, and [75] finds that the effects of global warming alone may cause
an increase of 95 megawatts in annual peak demand levels by 2030, largely due to more
consumption from air conditioners. Implementing measures to make these appliances more
efficient, and to make buildings better insulated, will lower total and peak demand levels.
This will have the effect of increasing the safety margin of generation capacity with existing
infrastructure.
Efficiency standards can also help stem the long-term growth in demand which will require large investments in new generation. Reducing the amount of new generation required
will make it easier to meet the 16% of total renewable generation mandated by the E.U.,
and [42] finds that "bringing the growth of consumption under control is crucial" to meet
those E.U. requirements.
35
2.4.3
Automated Energy Management
Efficiency is only one necessary component of energy demand policy in Cyprus, since a
recent study found the rise in household appliances in Cyprus has decreased the potential
to respond to prices or other signals to modify consumption patterns. The existing set
of efficiency subsidies could be augmented to include new standards requiring (or at least
subsidies to encourage) installation of automated energy management devices.
This is a
broad category of technologies which could include anything from thermostats that can be
scheduled to turn off at certain hours to the more advanced integrated management systems.
Much recent work has explored technological systems to enable automated communication and transactions between suppliers and consumers, and automated responsiveness of
demand to signaling. For example, in [41], Daniel Livengood and Richard Larson explored
a household energy management system, dubbed the Energy Box, which could act as a
proxy for the house's inhabitants. The inhabitants could input their personal temperature
preferences, for example, and the Energy Box could automatically react to price signals by
changing the thermostat's set point. In [40], Woei Ling Leow extends the work to include
per-room automation of thermostats, automatic learning of individual occupancy patterns,
and remote updating via smartphones. There are also various real-world building energy
management systems currently in use, with varying levels of usability and "intelligence".
These technological developments are promising, since they enable dynamic pricing
schemes without requiring constant consumer attention. Viewing the problem from a behavioral science standpoint, more advanced energy management systems make "passive
consumption", such as simple thermostats with little direct human intervention, relatively
more "active" on the part of the consumer by incorporating their preferences. As far back
as 1983, at least one study, [27], noted a general need for systems and policies favoring less
passive energy consumption patterns.
These devices could be one approach to reducing peak demand levels (which generation
must be built around) and, more broadly, enabling energy demand to be more responsive
to the dynamic state of the grid. These building management technologies don't necessarily
need to be rolled out to entire sectors at once; in fact, if they are introduced to a smaller
number of households first, lessons will be learned as their usage spreads. One option could
be to offer a subsidy for all houses, but a stronger subsidy or even a mandate to new building
36
construction.
2.5
Viability of Policy Approaches
Each of these approaches has been attempted in various electricity markets around the
world, often as small-scale pilot programs to begin, which allows for learning and improvement via experimentation. Cyprus's own coarse two-level TOU pricing options likely have
provided lessons from which Cyprus can learn, although little data on enrollment has been
published to date. Pricing the program well is vital, particularly in terms of how drastic
pricing differences are hour-to-hour and in terms of the payments given for responding to
signals to curtail or shift load. For example, if subsidies for participating in a load curtailment program are too low, then the reduction in total system costs caused by the program
participants is not fully internalized by the program, and there is a "free rider" effect due to
those who opt to keep flat rate pricing. On the other hand, if pricing variations from hour
to hour are too high, consumers could conceivably all increase or decrease usage in unison,
which can cause system instability.
[75] examines 50 years of electricity demand data in Cyprus and describes the behavioral aspects of the island's residential and commercial consumption. It finds that demand
has historically been mostly dependent on weather and outside temperature, and that large
price fluctuations may have small influence on electricity consumption. At first glance, this
seems to suggest that a dynamic pricing scheme is likely to be ineffective.
However, the
consumers analyzed in the 2009 study did not have the ability to see price fluctuations, nor
did they have devices to automatically adjust electricity usage of appliances, so the finding
is not directly indicative of dynamic pricing's potential. In fact, a U.S. Federal Energy Regulatory Commission analysis of a demand response program in Washington State provided
in [14] finds that program participants (who manually adjusted their power usage, without
automated devices) were initially overly eager to adjust their load in response to even very
weak price signaling. Their monthly savings were too small to justify the inconvenience, and
they quickly grew tired of the program. Finding a happy medium between weak and strong
price signals is challenging, and can determine whether a responsive demand program is
ultimately worth the effort, but it is possible.
Any policy implemented to make demand more responsive has the potential to lower
37
generation costs, curb pollution, and increase overall welfare. However, virtually any policyin this context or elsewhere-is susceptible to capture by affected parties, as described by
Stigler in [73]. For example, EAC [8] argues that since electricity utilities' profits rise as more
energy is consumed at more expensive, inefficient hours, they are likely to favor demand-side
programs that have a strong appearance of efficiency and sustainability but a low degree
of actual efficacy. In the case of Cyprus, it may help that the TSO is independent and designed to lower total system costs; however, the political influence of EAC, as the owner of
all electricity infrastructure, may outweigh all other concerned parties. Various new players
may enter the electricity sector in Cyprus, including manufacturers of building management
technologies and telecommunications providers responsible for enabling communication between homes and the TSO. If policies are designed in a way that requires the government
selecting winners, the potential for Stiglerian capture and rent-seeking exists. There are
mitigations for these risks, such as open bidding processes for new infrastructure work and
more generally a transparent policymaking process, but they are not foolproof.
A broader political failure problem associated with new responsive demand programs
is that of collective action. As described by Olson in [53], collective action problems arise
when the costs of an action or policy are concentrated but the benefits are dispersed. There
is not any single institution on the ground in Cyprus which would benefit greatly from more
responsive demand. Rather, the benefits would be dispersed among the populace in the form
of somewhat lower electricity bills and a more stable grid. Put another way, the options
are to continue the status quo-building more generation capacity owned and operated by
the utility, thus increasing its total profits-or to implement demand-side policies and build
slightly less new capacity. The latter option represents a sacrifice on the part of EAC, and
benefits that are widely dispersed. In the absence of a top-down impetus from the Cypriot
government or EU, it would require political support, and organization, of the populace
at large to advocate for those dispersed benefits, which Olson considers quite difficult. If
the case can be made to the public that demand-side investment pays for itself by lowering
the high cost of electricity in Cyprus, then perhaps public support and advocacy for such
investments can grow.
Even if these policies are open to capture, and difficult to even implement due to collective actions issues, they are still worth exploring. There is a wide body of evidence showing
the potential economic value of responsive demand programs, particularly on islands such as
38
Cyprus, and the technology exists. While real-time pricing may be the most optimal policy
in terms of absolute efficiency, time-of-use pricing or load curtailment incentive programs
would also increase the grid's ability to balance supply and demand. If these responsive
demand programs are deemed infeasible, simply extending existing efficiency subsidies and
standards to include automated energy management devices would lay the groundwork for
future demand-side policy.
Given that supply of electricity in Cyprus is far from fully
predictable and controllable, and that demand is increasing at an unsustainable rate, the
incentive to start assessing responsive demand policies sooner rather than later is growing.
39
40
Chapter 3
Modeling Framework
This chapter presents the model used to explore the value of responsive loads as well as the
data used to represent the Cypriot power system. The model demonstrates the changes in
hourly unit commitment scheduling allowed by the event-based demand response programs
described in Chapter 2.
The Wind Integration in Liberalized Markets (WILMAR) unit
commitment model is the basis of this model, which adds new parameters and constraints
for modeling responsive loads to WILMAR.
3.1
WILMAR Model Overview
WILMAR is a linear unit commitment optimization model developed by Riso National
Laboratory for Sustainability and the University of Stuttgart. WILMAR is intended to
evaluate the operational impacts of high penetration of wind power.
It determines the
least-cost (or highest-profit) level at which each generator in a power system should be
operated at a given time-known as a power system's unit commitment-using an hourly
time step.
WILMAR's unit commitment includes wind forecasting uncertainty through the use of
a rolling optimization time horizon. Rather than optimize over an entire modeling period
at once, it re-schedules unit commitment decisions in a series of optimization periods, each
starting an additional three hours further into the modeled scenario. A given optimization
period schedules generator output levels from the period's start time until midnight of
that day, and optimization periods that start at 12 PM or later extend to midnight of the
following day.
41
This iterative approach allows each optimization period to incorporate updated forecasts
for wind power. A given optimization period includes as input a scenario tree of 10 wind
forecasts, each with its own wind generation level and probability of occurrence. In that
optimization period, the generator output levels must be scheduled in a manner such that
demand and supply can stay balanced in all 10 scenarios. The scenario tree of wind forecasts
is updated for each optimization period, and the unit commitment schedule is re-optimized
each time. WILMAR's model structure and use of wind forecast error are explained in
greater detail in [62, 55, 21]. WILMAR has been used for several studies regarding high
penetration of wind power, including [28, 81, 43, 78, 26].
3.2
Approach to Modeling Responsive Loads
This model is similar to [23, 21] in that responsive loads are modeled as a form of energy
storage, which implies the shifting of load from one hour to another. It departs from that
approach by modeling load shifts as discrete events which are more tightly constrained.
These events are composed of a phase of load reduction from the status quo, starting at a
specific time, coupled with a corresponding phase of load increase from the status quo; each
of these phases is constrained to match more closely the behavior of a specific end use. It also
differs from that approach in that this model also includes loads in the commercial sector
which can be curtailed with no corresponding increase from the status quo. This curtailment
option is based very loosely on the Hawaiian Electric Company's demand response initiative,
described at [15]. It includes loads that can be purely curtailed (e.g., lighting, elevators and
signage) as well as some shiftable loads.
Figure 3-1 portrays an example of the kind of load shift event modeled here. In this
example, a dishwasher's unmodified "status quo" consumption is delayed from the 5PM
and 6PM hours to the 6PM and 7PM hours. This delay results in a net reduction in load
from the status quo during the 5PM hour and a net increase in load during the 7PM hour.
Because the dishwasher would run at 6PM with or without the shift, that hour's demand
levels are unchanged by the shift.
Reduction of the load reduces the available capacity
of the demand response program until after the corresponding load increase, or shifted
consumption, is complete. This essentially ensures that a given status quo consumption of
42
Status Quo Load
2
MW1
0
4PM
5PM
6PM
7PM
8PM
9PM
10PM
11PM
9PM
10PM
11PM
Shifted Load
2
MW1
0
4PM
5PM
6PM
7PM
8PM
Load Shift's Impact
2
1
MW
"
0
-1
Net Load
Increase
Net Load
Reduction
-2
4PM
5PM
6PM
7PM
8PM
9PM
10PM
11PM
Remaining Demand Response Capacity
4
3
MWh
2
1
0
4PM
5PM
6PM
7PM
8PM
9PM
10PM
11PM
Figure 3-1: A Sample Dishwasher Load Shift
a given participant cannot be curtailed twice in the model.1
Figure 3-2 portrays another kind of load shift, in which a thermal load is used. Two
megawatts worth of HVAC load is reduced during the 7PM hour, and spread across the
hours from 5PM-9PM. There are several fundamental differences between this load shift
and the dishwasher example:
" The HVAC usage is shifted in both directions from the status quo load, while the
dishwasher is simply delayed. Certain loads in the model is constrained to be shifted
only in certain directions, and each load has an upper limit on the amount of time for
which it can be shifted.
" In the HVAC case, the total energy of increased loads is slightly higher than the total
energy of reduced loads. This is because thermal load shifts incur a recovery penalty
from losses during the loading of thermal energy storage.
" The increased HVAC usage is spread over a longer duration than the reduced sta'Figures 3-1 and 3-2 do not reflect the minimum rest interval between demand response events that
restricts each demand response program. In the illustrations, if the program has a minimum interval of 24
hours between demand response events, then the capacity provided by the participants who shifted their
consumption remains offline for those 24 hours.
43
Status Quo Load
MW1
I
4PM
5PM
6PM
7PM
8PM
9PM
10PM
11PM
9PM
10PM
11PM
Shifted Load
2 -
MW 1
0
4PM
5PM
6PM
7PM
8PM
Load Shift's Impact
2
MNet Load
1
MW
0
MMIncrease
MNet Load
-1
Reduction
-2
4PM
5PM
6PM
7PM
8PM
9PM
10PM
11PM
Remaining Demand Response Capacity
4
3
MWh
2
1
4PM
5PM
6PM
7PM
8PM
9PM
10PM
11PM
Figure 3-2: A Sample HVAC Load Shift
tus quo usage.
Certain loads have a variable duration, while others (such as the
dishwasher) are constrained to operate for a fixed length of time.
The shifted HVAC consumption in this example can be interpreted as operating each participant's load at a lower hourly output level for a longer time, as if the HVAC unit cycles on
and off every 15 minutes rather than running constantly. Alternately, since this model optimizes aggregations of loads, it can be interpreted as distributing load users across multiple
time steps. In this example, that would imply that the consumption of 25% of participants
in this event are shifted to each of the four relevant hours (5PM, 6PM, 8PM, and 9PM).
The illustrated HVAC example portrays a constant magnitude of load increase during
the hours of thermal pre-conditioning and recovery. However, it is also possible for the
shifted load to be distributed more heavily on certain hours than others depending on what
is most optimal (for example, 6PM could gain twice as much increased load as the other
hours, or it could even be shifted entirely to 5PM).
44
3.2.1
Load Parameter Definitions
A combination of new and pre-existing load parameters are included in the model. The
following parameters are included in other versions of the WILMAR model for the modeling
of storage, and are reused here for responsive loads.
Demand response price: The energy payments (46 per MWh of shifted or
curtailed load) that must be paid in order to modify consumption patterns.
Minimum lead time: The number of hours of advance notice required to
schedule a load reduction or increase, or to cancel a scheduled reduction or
increase.
Energy capacity: The total energy (MWh) of the loads included in a program;
this is dependent on an end use's instantaneous power consumption as well
as its duration.
Power capacity: Denoted in Section 3.2.2 as C,, this is the maximum total instantaneous power consumption (MW) of the loads included in a program.
Recovery penalty: Denoted in Section 3.2.2 as RPp, this is the additional
energy consumed due to the shifting of a load, expressed as a proportion
of the load reduction.
The newly added load parameters are mostly focused on constraining the lengths and
offsets of each phase of a demand response event (load reduction and the corresponding
load increase). These parameters, along with the notation used in Section 3.2.2 to represent
them, are:
Minimum rest interval, minRestp: The minimum number of hours that must
pass after the completion of a load shift event prior to modifying those participants' consumption again
Minimum load duration, minLDp: The minimum number of hours allowed
between the start of a load reduction event and the end of the load reduction
(i.e., the minimum duration of the status quo load)
Maximum load duration, maxLDP: The maximum number of hours allowed
between the start of a load reduction event and the end of the load reduction
(i.e., the maximum duration of the status quo load)
45
Minimum shift, minSp: The minimum number of hours allowed between the
start of a load reduction event and the start of the corresponding load
increase. This value can be negative, which allows for advances in consumption, such as when thermal loads are pre-cooled or pre-heated.
Maximum shift, maxSp: The maximum number of hours allowed between the
start of a load reduction event and the start of the corresponding load
increase. This value is positive for all modeled loads (so there are no loads
which cannot be delayed rather than advanced).
Minimum shifted load duration, minSLDp: The minimum number of hours
allowed between the start of an event's load increase phase and its end.
Maximum shifted load duration, maxSLDp: The maximum number of hours
allowed between the start of an event's load increase phase and its end.
3.2.2
Load Shift Equations
Each of the load characterization parameters requires the addition of new constraint equations to WILMAR to define their impact on the optimization model's behavior.
These
equations are summarized here.
The model treats each demand response program as an aggregation of load of a given
end use (e.g., dishwashers). For a given program p, the program's overall impact on gridlevel demand at a given time t, changept, is defined by Equation 3.1 as the load increases
minus load reductions from the status quo of each participant.
change,,t =
inc,t,, -
redp,t,,
Vp, t
(3.1)
Each demand response program p in the model has a fixed size, Cp. This program size
is defined as a participation level-shared by all programs for a given modeled scenariomultiplied by the theoretical maximum capacity shown in Table 4.1. The program size sets
an upper limit on the magnitude of load reductions or increases from the status quo possible
at time t. The total magnitude of load reductions (or increases) from program p at time t
is equal to the sum of reductions (or increases) at t associated with load reduction events
starting at any time step r. Equations 3.2 and 3.3 express this relationship.
46
Indices
A program aggregating a given type of load
An hourly time step
The starting hour of a load reduction event being referenced
A past or future hourly time step at which a
load reduction event starts
Parameters
p
t, ti, or t 2
Ir
%-r
CP
RPp
minRestp
minLDp
maxLD,
minSp
maxSp
minSLDp
maxSLDp
changept
redStartp,t
redEndp,t,,
redp,t,r
incStartp,t2,
incEndp,t,,
incp,t,r
Program power capacity (MW)
Recovery penalty (unitless)
Minimum rest interval between load reduction events (hours)
Minimum load duration (hours)
Maximum load duration (hours)
Minimum shift (hours)
Maximum shift (hours)
Minimum shifted load duration (hours)
Maximum shifted load duration (hours)
Decision Variables (All Units in MW)
Overall change in load at time t due to program p
Magnitude of a load reduction event by p starting at t
Load reductions ended by p at t due to redStartp,,
Total load reduced by p at t due to redStart ,Load increases started by p at t due to redStart,
Load increases ended by p at t due to redStart,,
Total load increased by p at t due to redStart,,
Table 3.1: Summary of Notation
red,t, < Cp Vp, t, r
(3.2)
incp,t,, < Cp
(3.3)
Vp, t, -r
'
Each demand response event includes a load reduction event from the status quo, starting
at time r, and a corresponding load increase. Equations 3.4-3.7 define the possible duration
of the load reductions. Equation 3.4 constrains the magnitude of load reductions at time t
to equal the sum of all load reductions at time t - 1 and a new load reduction event that
may begin at t, minus any previous load reduction events ending at t.
redp,t,, =
redp,t-_,, + redStartp,t -
redEndp,,,
Vp, t
(3.4)
After a load reduction event starts, the event's magnitude cannot be decreased for at
least minLDp hours. This is the minimum load duration constraint, which Equation 3.5
47
defines using the running summation of all historical load reduction starts and ends.
ti -minLDP
t1
S
t2=0
redEnd(g, t 2 ) <;
E
t2=0
redStart(g,t 2 ) Vp, ti
(3.5)
Equation 3.6 uses a similar approach to define the maximum load duration constraint,
which prevents any load reduction event from lasting more than maxLDp hours.
ti -max
t1
redEnd(g,t2) ;>
o2=0
LD,
E
t 2 =0
redStart(g,t 2 ) Vp, t1
(3.6)
Equation 3.7 simply ensures that decreases in the magnitude of a load reduction event
cannot occur before minLD, hours have passed or after maxLDp hours have passed.
redEndp,t,, = 0 VT
s.t. T < t - maxLDp or r > t - minLDp
(3.7)
Equations 3.8-3.10 define the duration of the load increase portion of a demand response
event.
It is worth noting that some end uses, such as appliances, require that the load
increase (i.e., the shifted load) be of the same duration as the load reduction (i.e., the
status quo load).
However, others, including thermal loads and desalination units, can
have a different shifted load length within certain limits. As noted in Section 3.2, this can
be interpreted as operating a load at a lower output level for a longer time or as shifting
different participants to different hours.
Equation 3.8 defines the load increase at time t due to a load reduction event starting at
T, incpt,,, as the previous time step's value plus any additional increases due to the same
event, minus any increases that are now ending.
incp,t,r = incp,t-1,, + incStart,t,T - incEndp,t,,
Vp, t, T
(3.8)
This equation is similar in form to Equation 3.4; the main difference is that a different
summation is used because load increases begun at t may be due to multiple load reduction
events. For example, out of 3 MW of HVAC being brought back online at t, 1.5 MW may
be due to a load reduction starting two hours prior-i.e., a shift of 2 hours-while 1.5 MW
may have been due to a load reduction starting just one hour prior.
Equation 3.9 defines the minimum duration of load increases, minSLDp, in a manner
48
similar to Equation 3.5.
r-minSLDP
7-
S incEnd(g, ti, r)
<
ti=O
5
t2
incStart(g,t 2 , r)
Vp, ti,T
(3.9)
=O
Equation 3.10 defines the maximum duration of load increases, maxSLDp, in a manner
similar to Equation 3.6.
r-maxSLDp
incEnd(g, ti, r) ;>
incStart(g,t 2 ,T)
Vp, ti, r
(3.10)
Equation 3.11 defines the minimum number of hours a load can be shifted, minSp, as the
limit before which load increases cannot be greater than load reductions. This parameter,
minSp, is negative for certain loads, such as thermal loads which can be pre-conditioned.
incp,t,r < red(g, t -minSp,T)
1 + RPp
Vp, t, r
(3.11)
Equation 3.12 defines the maximum number of hours a load can be shifted, maxSp,
and requires that a shifted load fully recover reduced consumption within that interval
(accounting for differences between the allowed load durations and shifted load durations).
incp,t,' > red(g, t - (maxS -+maxSLDp - maxLDp),
r) Vp, t, -r
1+ RPp -
(3.12)
Equation 3.13 places an upper limit on the rate at which load reductions can be recovered; it can also be thought of as a lower limit on the time required to recover from a
load reduction. This constraint is only active on loads with maximum load durations above
one hour; it is intended to prevent unrealistically quick recoveries from load reductions of
loads which typically may operate near full capacity. For instance, a desalination plant may
operate near full capacity, and a load reduction event lasting two hours may require a load
increase for at least two hours to recover. 3.13
incpt'
redStart(g,T)
1 + RPp
Vp, t,T
(3.13)
Finally, Equation 3.14 defines the minimum rest interval between demand response calls,
minRestp, as the interval after a load reduction event ends before a given participant can
49
be called upon to reduce their load again.
tl
redStart(g,ti) < C, -
L
E redEnd(g,t2, T)
Vp,
ti
(3.14)
t2=ti-minRestp+1 'r
This interval is limited to a value no greater than 36 hours because WILMAR optimizes
only over the previous 24 hours, and for as few as 12 hours in advance. This is not reflective
of existing event-based demand response programs, which would rarely, if ever, request load
curtailments or shifts that frequently. However, in the modeled scenarios, this constraint is
rarely binding due to the operational costs of demand response.
3.3
Approach to Modeling CSP with Thermal Energy Storage
A simple model is used to represent concentrating power (CSP) with thermal energy storage.
Rather than building a new model from scratch, existing constraint equations for storage
devices and conventional generators are leveraged to construct a generator with a storage
capacity.
Solar Radiation
Hourly storage
charging data
Thermal Storage
- Max. capacity
- Hourly decay rate
Steam Generator
- Max. output
- Min. load factor
- Part-load efficiency
- Startup constraints
a Variable costs
- Min. lead time
Figure 3-3: CSP+TES Modeling Approach
As shown in Figure 3-3, the thermal energy storage is charged using fixed inputs from
solar irradiation data. It is not permitted in the model for this storage device to draw power
from the grid. As the storage is charged from the sun, a percentage is lost due to the hourly
decay rate of the thermal storage medium. This retains the storage's potential to provide
flexibility to the grid and to smoothen its output, but also incentivizes the model to use
solar resources to generate electricity immediately when all other factors is equal.
The thermal storage medium provides heat to power a steam generator. WILMAR's
existing constraints for thermal generation are reused for the stage of the CSP model,
including:
50
"
Maximum output level
" Minimum load factor
" Part-load and full-load efficiencies
" Variable O&M and ramping costs
* Minimum lead time required to schedule generation
" Minimum operation time
* Minimum shutdown time
These constraints are used without modification, and are described in [55].
The values
chosen for each CSP parameter, as well as other model parameters, are given in Chapter 4.
51
52
Chapter 4
Input Data
4.1
Load Parameter Values
In the modeled scenarios, demand response program size is defined as a percentage of the
theoretical maximum capacity of a program including a given end use. The method used
to estimate the theoretical maximum capacity for each type of end use differed by sector,
due to different data sources for each sector.
The theoretical maximum program capacities, maxCp, of residential end use programs
p are calculated using Equation 4.1:
maxCp = households x ownershipRate x deviceConsumption x availability
Where the variables are defined as:
households: The total number of households in Cyprus, assumed to be 223,790
per [63], the most recent national census available online.
This value is
scaled to match projected population levels in 2020 for the modeled scenarios.
ownershipRate: The average number of devices per household, mostly adapted
from Limassol ownership data found in [38].
deviceConsumption: Typical electricity consumption due to a given end use,
described in [47], a fact sheet prepared by EAC to educate consumers on
potential energy savings.
53
(4.1)
availability: An adjustment factor used to account for the fact that certain end
uses-namely, appliances-are unlikely to be in use all day, meaning there
may be no active "status quo" load which can be shifted. Due to a lack
of suitable data for this factor, a value of 0.1 for all shiftable appliances is
used. This may overestimate the availability of certain appliances such as
laundry machines, which are used for less than 10% of the time. However,
this bias is lessened somewhat by the fact that appliance usage is generally
more likely to occur during hours of high demand (which are the hours in
which demand response may be more likely to be utilized).
The estimated maximum capacities of each modeled end use if every household participates
in demand response are shown in Table 4.1. Program capacities for each modeled scenario
are scaled using these values; for instance, for a scenario in which demand response program
participation levels are at 50%, the program capacity of residential HVAC is 6.1 MW.
Theoretical
Maximum
Consumption
Per Device
(kW/hr)
Ownership
Rate
Hourly
Availability
Factor
Capacity (MW)
Residential HVAC
0.09
60.0%
100%
12.2
Res. Refrigerators
Res. Freeezers
0.09
0.11
100.0%
50.0%
100%
100%
20.1
12.3
Res. Dishwashers
1.05
71.5%
10%
16.8
Res. Washers
1.02
44.1%
10%
10.1
Res. Washer-Dryers
Commercial HVAC
2.08
N/A
55.9%
N/A
10%
N/A
26.0
47.6
Comm. Curtailment
N/A
N/A
N/A
11.9
Desalination
Total
N/A
N/A
N/A
11.9
168.9
Table 4.1: Estimated Theoretical Maximum Responsive Load at 2012 Demand Levels
Commercial and industrial loads are not included in [38]. Instead of using Equation 4.1,
their maximum capacities are approximated using hourly average aggregate commercial
consumption data from [37].
This aggregate is assumed to be composed of 75% uncon-
trollable loads, 20% HVAC loads, and 5% curtailable loads with no shift necessary, such as
lighting, elevators, and signage. Due to a lack of granular data on the types and magnitudes
of industrial and public works loads in Cyprus, or on which industrial consumers may be
0
Hawaiian Electric Company's growing commercial direct load control program is a rough reference point.
In 2011, it included roughly 40 commercial customers and 20 MW of load. Hawaiian Electric Company's
system, , with an installed generating capacity of roughly 1900 MW, is slightly larger than Cyprus.[16, 17]
54
willing to participate in demand response, only one such end use is included: desalination,
which is simply assumed to have a maximum capacity equal to commercial curtailments.
Other operational constraints, including allowed load durations, lead time required to
initiate a demand response event, and allowed times across which a load can be shifted, are
shown in Table 4.2.
Min.
Min.
Max.
Lead
Time
Load
Duration
Load
Duration
Min.
Shift
Max.
Shift
Residential HVAC
0
1
3
-3
3
Res.
Res.
Res.
Res.
Res.
0
0
3
3
3
1
1
2
2
3
1
1
2
2
3
-2
-2
1
1
1
2
2
12
3
3
Commercial HVAC
0
1
3
-3
3
Comm. Curtailment
Desalination
0
0
1
1
2
4
N/A
-4
N/A
4
Refrigerators
Freeezers
Dishwashers
Washers
Washer-Dryers
Table 4.2: Modeled Operational Constraints of Responsive Loads (All Units in Hours)
Aside from minimum and maximum load durations, which are taken from [47], these
operational parameters are chosen to approximate each end use's likely characteristics. End
uses which could be automated relatively easily (e.g., thermal loads) are given a lead time of
less than one hour. The same lead time is given to end uses in the commercial and industrial
sectors, which may have high enough consumption to warrant installation of special direct
load control equipment. On the other hand, residential appliances such as laundry machines
and dishwashers may be more difficult to automate-since the appliances are not ready to be
run until loaded by users-so they are assigned a longer lead time. Allowable shift lengths
are derived loosely from a Portuguese case study on end user's willingness to have loads
controlled, found in [70]; loads that can be shifted with relatively little loss of comfort
are allowed longer load shifts. For desalination plants, [29] is used as a starting point for
assigning operational constraints. In this case, the model assumes the desalination plant
must meet its daily production goals, but up to four hours per day can be shifted for up to
four hours in either direction.
Table 4.3 lists the cost components assigned to each end use. Relative values for willingness to modify consumption for residential end uses (on a scale from 0 to 1, where 1 is
most willing) are adapted from [70]; they are used as a proxy for operational costs, since
55
Residential HVAC
Res. Refrigerators
Res. Freeezers
Relative Willingness
to Modify Consumption
0.5
0.4
0.4
Cost
Multiplier
1.167
1.273
1.273
Recovery
Penalty
25.0%
5.3%
5.3%
Res. Dishwashers
Res. Washers
1.0
0.8
0.824
0.933
N/A
N/A
Res. Washer-Dryers
0.8
0.933
N/A
Commercial HVAC
0.4
1.273
11.1%
Comm. Curtailment
Desalination
0.2
0.2
1.556
1.556
N/A
11.1%
Table 4.3: Modeled Cost Components of Responsive Load Shifts or Curtailments
an ideal demand response program would compensate participants at a level corresponding
to their loss of comfort. Commercial HVAC consumption patterns are assumed slightly less
modifiable than for residential HVAC due to the tourism industry's emphasis on customer
satisfaction. Commercial curtailments are assumed half as modifiable as commercial HVAC,
since curtailment represents a loss of welfare which is not made up by a shift; desalination
is simply assumed to be as costly as curtailments.
The cost multiplier is the ratio of the midpoint of the values from [70]-0.7-to the value
assigned to each end use. The multiplier is applied to the modeled price of demand response
in each scenario. For example, in a scenario with a stated demand response price of 50% the
cost of a peaking unit, the price to use desalination for demand response would be 77.8%
the cost of a peaking unit.
The recovery penalty, RPp, used in Equations 3.11 - 3.13, applies only to certain loads.
It is defined as
I
-
1. As a very rough approximation for modeling purposes,
refrigerators and freezers are assumed to have a storage loading loss of 5%. Buildings, on
the other hand, are assumed less efficient at retaining thermal conditioning, and given a
higher storage loading loss: 10% for commercial buildings and 20% for residences, which
are assumed to be even less well-insulated. The recovery penalty for desalination loads does
not represent the loss of thermal storage, but rather the loss in process efficiency compared
to the status quo scheduling of desalination.
56
4.2
Conventional Generation Parameter Values
Certain operational parameters for each conventional power plant in Cyprus are provided
by Dr. Constantinos Varnavas, Assistant Manager at the Generation Unit at the Electricity
Authority of Cyprus. These parameters include each generating unit's capacity, minimum
load factor, minimum shutdown time ("downtime"), and minimum operation time ("uptime"). The operational parameters used in the model are given in 4.5.1
#
Fuel
Capacity Per
Unit (MW)
Dhekelia ST
Vasilikos ST
Dhekelia ICE
Moni OCGT
HFO
HFO
HFO
LFO
60.0
130.0
17.0
37.7
6
3
6
4
360.0
390.0
102.0
150.8
Vasilikos OCGT
Vasilikos CCGT 1
LFO
LFO
72.5
110.0
1
1
72.5
110.0
Vasilikos CCGT 2
Total
LFO
220.0
1
22
220.0
1405.3
of
Units
Total
Capacity (MW)
Table 4.4: Overview of Modeled Conventional Generators
Minimum
Load Factor
Lead Time
(hrs)
Minimum
Downtime (hrs)
Minimum
Uptime (hrs)
Dhekelia ST
Vasilikos ST
Dhekelia ICE
0.50
0.46
0.58
4
4
0
8
8
1
8
8
2
Moni OCGT
Vasilikos OCGT
0.11
0.11
0
0
1
1
3
3
Vasilikos CCGT 1
Vasiliikos CCGT 2
0.55
0.55
4
4
6
6
8
8
Table 4.5: Modeled Operational Constraints of Conventional Generators
Certain parameters related to generator startup and shutdown are adjusted from the
provided values to better work within the WILMAR model's linear constraints. For example, to maintain linearity, the WILMAR model does not model generator startups as an
integer (on/off) decision variable, but rather as a continuous variable.
This changes the
meaning of the minimum uptime from "the time a generator must remain on" to "the time a
'The Vasilikos CCGT 1 unit is modeled as two separate units: one combined cycle gas turbine unit which
includes a gas turbine and a heat recovery steam generator, and a second, independent unit including only
one gas turbine. This configuration is sometimes used to increase the flexibility of combined cycle units with
regard to ramping rates and minimum output levels. However, the full-load efficiency is lower than if the
second gas turbine were also connected to the steam generator.
57
generator's output level, once raised to a given value, must remain at or above that value".
Similarly, minimum load factor is not a proportion of a unit's installed capacity, but rather
a proportion of the capacity that is considered online at a given time, which can take on any
value up to the actual installed capacity. 2 Because of this discrepancy, the values provided
by EAC inform the parameter values listed in Tables 4.4 and 4.5, but are not all directly
input to the model.
Dhekelia ST
Vasilikos ST
Dhekelia ICE 1-3
Dhekelia ICE 4-6
Moni OCGT
Vasilikos OCGT
Vasilikos CCGT 1
Vasiliikos CCGT 2
Part-Load
Full-Load
O&M Costs
Ramping Costs
Efficiency
0.29
0.37
0.44
0.43
0.10
0.10
0.42
0.42
Efficiency
0.32
0.40
0.44
0.45
0.30
0.30
0.47
0.48
(F /MWh)
1.5
1.5
2.5
2.5
6.0
6.0
3.2
3.20
(F /MW)
40.0
40.0
8.0
8.0
80.0
80.0
20.0
20.0
Table 4.6: Modeled Variable Cost Components of Conventional Generators
The WILMAR model also requires certain financial data be input for each generating
unit, including operations and maintenance (O&M) costs and the cost to increase a turbine's
output level. The modeled values are shown in Table 4.6. Approximations for these values
are derived from a several sources: two National Renewable Energy Laboratory reports
([44, 64]), the values packaged within the WILMAR model's sample datasets for similar
generator types, and general knowledge of each generator group's current usage.
One of the largest and least predictable cost components of electricity generation is
the cost of fuel. Monthly average costs of fuel oil used for generation are available on the
EAC website, and projections for 2020 fuel prices are developed using a combination of
International Energy Agency assumptions and projections from [6, 4, 5] and U.S. Energy
Information Administration estimates from [2]. As shown in Table 4.7, the cost of fuel oil is
assumed to remain similar to recent years (which follows a projected decline and rebound
in prices in the interim period). Natural gas is projected to cost just below 75% as much
as heavy fuel oil, and just over 50% as much as the more refined #2 diesel oil.
2
These are fair approximations when modeling large numbers of generators as a single unit, but they
limits the model's fidelity when representing individual turbines. Still, when spinning reserve capacity must
be maximized, online capacity should equal the installed capacity, which makes the minimum load factor
approximation more accurate.
58
Fuel Type
Price (4 /GJ)
Price (4 /MWh)
12.47
17.13
9.04
44.89
61.68
32.55
Heavy Fuel Oil
#2 Diesel Oil/Gas Oil
Natural Gas
Table 4.7: Modeled Fuel Prices
Because demand response is scaled according to the variable costs of conventional generating units, which include fuel costs, the model's value assessment of demand response
is expected to be relatively insensitive to the absolute values of fuel prices. More notable
impacts could result from price differences between each fuel type, which affect the merit
order of conventional generating units.
4.3
CSP Generation Parameter Values
The assignment of operational and cost parameters to CSP generation is more speculative,
because none of the modeled CSP generators are operating yet and CSP technologies may
evolve before they are built. Four CSP projects are included in the model for 2020, as shown
in Table 4.8:
" A 50 MW trough system to be located near Akrotiri, planned by EAC and described
in [46, 60]. The system is assumed to have enough storage capacity for five hours of
full-load operation in this model.
" A 4.5 MW pilot project at Pentakomo, sponsored by the Cyprus Institute, to cogenerate desalinated water and electricity in an integrated manner, including storage for
up to 24 hours. This project would use a novel design for a central collector storage
medium, and is further detailed in [7, 58, 36].
" A planned 50.76 megawatt Stirling dish system with no storage capacity recently
awarded funding by the EU's NER300 Program, as announced in [20].
" A similar 25.5 megawatt Stirling dish system with no storage capacity to be built
outside Nicosia, as described in [54].
Reference sheets including [3] exist for operational parameters of steam turbines that
may be used in dispatchable CSP plants such as the modeled ones at Akrotiri and Pentakomo. However, due to the uncertainty surrounding technological development of CSP
59
Akrotiri Trough
Pentakomo CSP-DSW
Larnaca Stirling
Nicosia Stirling
Total
Generation
Capacity (MW)
50.0
4.5
50.8
25.5
130.8
Hours of
Storage
5
24
0
0
Storage
Capacity (MWh)
250.0
108.0
0.0
0.0
358.0
Table 4.8: Overview of Modeled CSP Generators
between now and 2020 and the wide range of possible technologies for these projects, the
startup parameters in Tables 4.9 and 4.10 are chosen to make the CSP units slightly more
flexible than modeled CCGT units, and with O&M costs similar to modeled steam units.
Akrotiri Trough
Pentakomo CSP-DSW
Minimum
Load Factor
Lead Time
(hrs)
Minimum
Downtime (hrs)
Minimum
Uptime (hrs)
0.4
0.4
4
4
3
3
3
3
Table 4.9: Modeled Operational Constraints of CSP Generators
Akrotiri Trough
Pentakomo CSP-DSW
Part-Load
Efficiency
Penalty
8.5%
8.5%
O&M Costs
Ramping Costs
Storage
Decay Rate
(E /MWh)
1.5
1.5
(s /MW)
20.0
20.0
(%/hour)
5.0%
5.0%
Table 4.10: Modeled Variable Cost Components of CSP Generators
4.4
Solar Resource Data
Solar resource data is provided by Dr. Alexandros G. Charalambides at the Department
of Environmental Science and Technology of the Cyprus University of Technology. The
data provided includes global irradiance, direct normal irradiance, and temperatures for
Cyprus in one-minute increments, which are converted to hourly average values for use in
the WILMAR model. The temperature at a given hour is used to adjust the PV potential
to account for reduced operating efficiencies of photovoltaic panels at high temperature.
Direct normal irradiance data is used as the input for CSP generators.
Two days worth of data per month-one typical day, and one day with maximum sunlight
for that month-are used to generate data for the full year. The algorithm described in B
60
converts that data into a full year of solar resource input data. The algorithm is tuned to
yield annual solar capacity factors roughly equal to the projected capacity factors from the
Cyprus National Renewable Energy Action Plan.
Installed
Capacity
Annual
Minimum
Annal
Maximum
Daytime
Median
Daytime
Mean
Annual
Capacity Factor
192.0
0.0
155.5
65.2
65.5
19.0%
Table 4.11: Input Data Statistics for PV. Units are in MW.
Table 4.11 gives several statistics regarding the PV input data used.
The installed
capacity is never fully realized. This is partially due to the relatively coarse hourly timescale
used; the peak daily global irradiance may occur for only several minutes, which is not
captured well when using hourly averages.
It is also due to the seasonality of the PV
input data. The maximum peak input values occur in the summer months, which are also
the hottest months, with temperatures peaking at 36.5 C (97.7 F) on August 9. At that
temperature, the model assumes the efficiency of PV panels is degraded, as described in B.
Installed
Capacity
Annual
Minimum
Annal
Maximum
Daytime
Median
Daytime
Mean
Annual
Capacity Factor
76.3
0.0
75.5
43.2
38.5
28.1%
Table 4.12: Input Data Statistics for CSP Without Storage. Units are in MW.
Table 4.12 describes CSP input values for CSP without storage. For those units, the
same method is used as PV to scale the direct normal irradiance input data. In this case,
the installed capacity of CSP is nearly fully realized. This is because the peak daily direct
normal irradiance occurs over a somewhat more gradual time scale, and is converted to
hourly averages with relatively little loss of peak value. It is also because CSP performance
is not negatively impacted by outside temperature in this model.
Values are not shown in Table 4.12 for CSP equipped with storage and steam generators
because the output is dispatchable, and the realized values depend on the optimal scheduling
of each unit.
For such units, the installed capacity is scaled to correspond to 5/8 the
annual peak direct normal irradiance. For example, a CSP unit with a maximum generation
capacity of 50 MW is capable of loading its thermal storage at a maximum rate of 80 MW
of equivalent electricity per hour on the day of the year with maximum irradiance. This
scaling value allows dispatchable CSP units to operate near full load during the clearest
61
hours of each day (as opposed to only during the clearest hours of the year), or to load
storage instead if that is more optimal.
Unlike wind, solar potential is assumed to be perfectly forecast. The high availability of
solar resources in Cyprus warrants exploring the effects of solar forecast error on the model.
However, this limitation is mitigated by the fact that the presence of sun is generally more
predictable than wind.
4.5
Wind Resource Data
The Cyprus TSO's website provides historical data for system-level wind generation in 15minute time steps. The hourly average of that data from the year 2012 is used, and scaled
linearly to match the planned wind capacity in 2020 (300 MW). To eliminate the effect of
new wind generation brought online during 2012, the data is normalized to the installed
capacity at a given time. There are days on which the historically measured data exceeds
the capacity that had been officially brought online per the TSO's website, so the installed
capacity on a given day is assumed to be the maximum of the official online capacity and
the historical running maximum wind production.
Installed
Capacity
Annual
Minimum
Annual
Maximum
Annual
Median
Annual
Mean
Annual
Capacity Factor
300.0
0.0
263.3
23.2
38.4
12.8%
Table 4.13: Wind Input Data Statistics. Units are in MW.
As shown in Table 4.13, this approach results in a capacity factor of 12.8%, a low value
which may underestimate the actual generation from wind that should be expected and the
resulting issues of intermittency (as one point of reference, [57] describes a capacity factor
of 15% in Cyprus). The modeled generation from wind is low enough that wind is rarely
curtailed in the 2020 scenarios, but high enough that intermittency leads to significantly
increased ramping of generators and sub-optimal operation due to the need for operating
reserves.
A wind scenario creation tool created by D. Karl Critz for [21] is used to create stochastic
wind data. The tool uses the scaled wind data as the realized production, but generates a
scenario tree of 10 randomly generated forecasts, each with a randomly generated probability
of occurrence. WILMAR's objective function optimizes over the weighted average of these
62
forecast scenarios, and the model is constrained such that demand and supply must remain
matched for all 10 possible scenarios.
4.6
Electricity Demand Data
The Cyprus TSO's website also provides historical system-level electricity demand data in
15-minute time steps, both for day-ahead forecasts and realized demand. Day-ahead forecast
data is used instead of actual data because there are fewer anomalies such as missing data
points. There is no clear bias between the two data sets in terms of absolute quantities
or hourly variability. Because the WILMAR model uses an hourly time step, the hourly
average of the 15-minute data is used.
Anual
Minimum
Annual
Maximum
Annual
Median
Annual
Mean
Annual
Load Factor
264.3
992.8
524.3
533.2
53.7%
Table 4.14: Demand Input Data Statistics (2012). Units are in MW.
Demand data from the year 2012 is scaled linearly to match the proper level of peak
demand for a given scenario. Table 4.14 shows statistics for the 2012 input data for demand.
Due to this approach, the input data does not reflect future changes in load factor or daily
load shape. For instance, the possibility that HVAC usage at peak hours could contribute
to a larger portion of overall energy consumption, thus reducing the system load factor and
requiring more daily generator ramping, is not captured in this model.
63
64
Chapter 5
Results from Modeled Scenarios
5.1
Overview of Scenario Design
The model described in Chapters 3 and 4 is run across a set of 24 scenarios with a range
of operational and economic parameters.
This range of parameters helps capture some
of the uncertainty surrounding projections of the Cypriot power system in the year 2020.
While the model used here may approximate the system's behavior, some of the strongest
determinants of a demand response program's value fall outside the scope of the optimization
model. Those determinants fall into three very broad categories: the state of the Cypriot
economy, the development of domestic natural gas resources for generation, and the design
of a demand response program.
Aside from various other economic impacts, the recent Cypriot financial crisis will affect
electricity demand, most likely as a short-term decrease in demand followed by a long-term
steady increase. The most TSO projections, published as they are updated at [51], forecast
2020 peak demand to fall somewhere between 975 and 1235 megawatts, revised down from
earlier projections as high as 1525 megawatts. To help include this uncertainty, two values
for 2020 peak demand are used to scale hourly demand levels:1
9 975 megawatts: this projection assumes that Cyprus endures a steady but gradual
economic recovery. It corresponds to a 2020 peak demand roughly equal to the year
2012's peak demand forecast (reached by taking the hourly average of historical data
'Because they most closely match current projections, scenarios with slow growth in demand are analyzed
in greater detail here. The impacts of faster demand growth are discussed, but the full set of figures for
scenarios with fast demand growth is included in Appendix C.
65
in 15-minute increments available on EAC's website).
* 1275 megawatts: this projection assumes that the Cyprus economy recovers more
quickly than current forecasts predict. It corresponds to a 31% increase in demand
over historical data from the year 2012.
Such a drastic gap will impact the need for future investment in generation capacity. For this
reason, the third 220 megawatt CCGT unit at Vasilikos, currently scheduled for construction
by 2015, is excluded from the model in the cases with lower peak demand.
As exploration of potential offshore resources proceeds, studies such as [1] are underway
to assess how to best utilize any discoveries, and there are a range of possibilities. With
regard to electricity generation, EAC plans to switch to natural gas-dominated generation,
but cost-benefit analyses would be necessary to assess which units should be converted to
natural gas, two scenarios are considered here:
" Natural gas generation:
most conventional generation units are converted to burn
natural gas instead of fuel oil. The exceptions are the steam turbines at Dhekelia and
all open cycle gas turbines. The Dhekelia steam units are among the system's oldest,
and the gas turbines are inefficient and mostly used for peak and backup capacity; for
this reason, investing in upgrades is less likely.
" Continued reliance on fuel oil: in this scenario, the current fuel mix is unchanged in
2020.
The latter scenario could occur if circumstances prevent or delay the switch to natural gas,
or if other uses of the natural gas resources are deemed more efficient.
The design details of programs to incentivize responsive demand also fall outside the
scope of this model, despite their significant impact on the programs' operational value. As
described in Chapter 2, various business models and policies exist for event-based demand
response or other forms of responsive demand. Following an approach similar to that of
[23], a range of program participation levels and values are simulated using this model:
" Program participation ranging from 30% to 70% of a rough estimate of the theoretical
maximum level of controllable demand in Cyprus.
" Price levels ranging from 30% to 70% of the variable costs of a peak unit operating
66
at full load. In the case of Cyprus, the benchmark price corresponds to an open cycle
gas turbine, with variable costs of S 0.21/kWh at full load.
The method used to calculate the demand response program electricity capacity of each
end use and the method used to calculate price levels for each end use are described in
Chapter 3.
The price levels modeled here are low relative to the values modeled in [23].
This
difference is used partly to account for different methodologies. This model mandates an
electricity consumption (including, in some cases, a recovery penalty) shortly before or
after any curtailment, depending on end use. This enforced rebound effect is partially offset
by a lower range of values, and it also makes direct comparison of price values between
models.
Furthermore, the open cycle gas turbine units used as a price benchmark are
inefficient enough that they are rarely, if ever, used for generation in the baseline model
runs described below, necessitating relatively lower modeled prices for demand response.
Natural Gas
9752 ?
0%
Fuel Oil
1275
975
1275
N/A
S
S+D
S
S+D
30%_
70%
S
S
D
D
S
D
S
D
50%
50%
S
S+D
S
S+D
70%
30%_
70%
S
S
D
D
S
S
D
30%
D
Table 5.1: Stochastic (S) and Deterministic (D) Cases
2
Table 5.1 summarizes the modeled cases. All cases listed model a full year of operation.
Two versions of the model are used-stochastic and deterministic-although all results in
this chapter use the stochastic model, and deterministic results are presented only in the
sensitivity analysis figures found in Chapter C.
The deterministic version does not include a tree of wind forecast scenarios, but rather
one fully accurate forecast. Because there is no forecast error in the deterministic version,
there is no need for generators to maintain the ability to ramp up or down output to
2
All results presented here are taken from stochastic cases except where labeled otherwise.
67
balance forecast error. This effectively means generation can be optimally scheduled even
without the flexibility provided by demand response.
However, benefits are still shown
simply because demand response is still cheaper than certain generating units, and it can
still reduce generator ramping by balancing renewable fluctuations. The primary benefit of
the deterministic mode is its faster runtime.
Because the cases including a gradual growth in demand match current projections
most closely, stochastic simulations are run on all of those cases. Stochastic simulations
are also used for all baseline cases and for all cases at the center point of demand response
program design parameters (50% price and 50% participation levels), to allow comparison
of modeled conventional generation and peak demand scenarios both with and without
demand response. Deterministic runs are also performed for all of these cases involving
fast demand growth, simply to allow sensitivity to demand response price and participation
levels to be compared directly to a deterministic baseline.
5.2
Results for 2020 with Natural Gas Generation and Gradual Demand Growth
5.2.1
Unit Commitment Scheduling
The Cyprus TSO predicts 2020 peak demand and annual energy consumption to be roughly
at the same level as at present, given the recent economic crisis, and Cyprus's current plan
is to leverage domestic natural gas to reduce the cost of fuel for power generation. This
scenario models those two outcomes. As at present, the relatively inefficient open cycle gas
turbines continue to burn #2 diesel oil (also referred to as gas oil or #2 distillate) and the
relatively old Dhekelia steam units continue to burn heavy fuel oil. All other conventional
generating units in this scenario burn natural gas.
The resulting load duration curves,
including the effect of wind, PV, and CSP (some of which is dispatchable), are shown in
Figure 5-1.
As noted in Section 4.2, the price of natural gas used in the model is 4 32.55/MWh.
Given the large gap between this price and the prices of imported heavy fuel oil and #2 diesel
oil-e 44.89/MWh and I 61.68/MWh, respectively-this effectively moves all gas-fired units
up in the merit order for unit commitment. Figure 5-2 shows the variable cost components
for each generating unit and demand response end use in this scenario; since conventional
68
1000
Demand
900
Demand minus Wind
800
Demand minus Wind and PV
Demand minus Wind, PV, and CSP
700
600
.0
500
400
300
200 100 0
0
1000
2000
3000
4000
5000
Hours
6000
7000
8000
Figure 5-1: Load Duration Curve
(2020, Natural Gas, Gradual Demand Growth, No Demand Response)
generation has a large fuel cost component, part- and full-load variable costs are each
indicated.
Another effect of the large price gap between natural gas-fired units and the other units
is to raise the relative price of demand response, since each option is priced relative to a
peaking gas turbine. In this scenario, every demand response option is more expensive than
any natural gas-fired unit, but also less expensive than the other units.
Aside from the switch to natural gas, this scenario (and each one modeled here) differs
from present-day Cyprus in that renewables account for a much larger share of generation.
The variability of wind and solar generation is apparent in Figure 5-3, which shows model
output from a typical week in January in the baseline (no demand response) scenario.
Wind generation does not surpass 30 MW-10% of installed capacity-from the morning
of January 26 until the evening of Wednesday, January 29 in this particular week. However,
the following morning, it rapidly rises to nearly 150 MW. In order to accommodate that
rapid increase, ICE units must ramp down, then ramp back up when wind generation
dies down. Intermittency and forecasting error of wind is also largely responsible for the
fluctuating output levels of the steam turbines, which handle most of the hourly balancing
of wind. Their role in providing balance is not due to faster or cheaper ramping ability-in
69
Larnaca CSP Stirling
Nicosia CSP Stirling
Solar PV
Wind
Pentakomo CSP DSW
Akrotiri CSP Trough
Vasilikos CCGT 2
Vasilikos CCGT 1
Dhekelia ICE 2-3
Dhekelia ICE 2-2
Dhekelia ICE 2-1
Dhekelia ICE 1-3
Dhekelia ICE 1-2
Dhekelia ICE 1-1
Vasilikos Steam 3
Vasilikos Steam 2
Vasilikos Steam 1
DR Res. Dishwasher
DR Res. Washer-Dryer
DR Res. Washer
DR Desal
DR Comm. HVAC
DR Res. HVAC
DR Res. Refrig-Freezer
DR Comm. Curtailments
Dhekelia Steam 6
Dhekelia Steam 5
Dhekelia Steam 4
Dhekelia Steam 3
Dhekelia Steam 2
Vasilikos CCGT 1 Peak
Vasilikos OCGT 1
Moni OCGT 4
Moni OCGT 3
Moni OCGT 2
Moni OCGT 1
-0.8
MW
Non-fuel variable costs
Fuel cost at maximum load
Extra fuel cost at minimum load
-25.5 MW
- 192.0 MW
-300.0 MW
4.
50.0
0
MW
MW
100
200
400
300
Variable costs (euros/MWh)
500
600
700
Figure 5-2: Variable Costs Per Unit
(2020, Natural Gas, Gradual Demand Growth, DR Price=50%, DR Participation=50%)
fact, their ramping costs are significantly higher than the ICE units-but due to the ICE
units' much lower operating costs.
CSP and solar PV are perfectly forecast in this model, but their variability still increases
the complexity of the unit commitment scheme. Because CSP requires direct irradiation, it
is affected significantly by cloudy days. This is reflected in its output, which reaches only
approximately 25 megawatts on January 27 then surpasses 100 megawatts the following
day. This affects the daily cycling schedules of conventional generation. Because the steam
turbines will be needed to provide power in the morning of the 27th, and they have a
minimum shutdown time of 12 hours, they must be left running overnight, so the ICE units
are cycled down. The following night, enough CSP and PV output is forecast to allow the
steam turbines to be cycled off, and the ICE units remain online overnight.
70
CCGT
ICE [fL
ST
CSP+TESE
]
CSP
PV
\\ind
W
1000
800
600
400
200
0
200
C )150
C,
100
C,
C
C,
50
I
50
C
50
/
/...
......
I.
0
100
800
600
400
200
0
12AM, 01/24
12AM, 01/25
12AM, 01/26
12AM, 01/27
12AM, 01/28
12AM, 01/29
12AM, 01/30
12AM, 01/31
Time
Figure 5-3: Unit Commitment in Baseline Case, 1/24-1/30
(2020, Natural Gas, Gradual Demand Growth, DR Price=oo, DR Participation=0%)
While solar and wind add to the complexity of the unit commitment scheme, CSP with
thermal energy storage also adds an extra option for flexibility. The model assumes an
hourly storage decay rate of 5%, incentivizing the CSP units to produce at as high a level
as possible, but there is still a benefit to deferring generation on certain days. This can be
seen in Figure 5-3 by comparing the output of CSP without storage to that of CSP with
storage; the two options have identical resource input shapes. On January 27, generating
power instantly in the morning would simply force steam units to ramp down, so storage is
used to fully shift generation to the afternoon peak demand hours. On the other hand, on
January 29-another cloudy day-generating power instantly in the morning allows startup
of the steam units to be deferred, and the storage is utilized much less.
Demand response is another source of flexibility. Figure 5-5 illustrates the same time
71
period in January with the addition of demand response, using a participation level of 50%
and a price of 50% the full-load cost per megawatt-hour of a peaking gas turbine. The
area chart colored red gives the total net generation (positive) or consumption (negative) of
demand response programs in a given hour. A positive value corresponds to a net reduction
in consumption versus the status quo at that hour, while a negative value corresponds to a
net increase in consumption.
CCGT
ICE
ST
CSP+TES
CSP [
PV
Wind
DR
1000
400
200
0
5.
A,
-5
200
50
0
. . . . . . . . . . . .. . .
......
. . . .. .. . .... . .
a)
.......
C)
0
C)
150
0
100
.. ..
0
150
50
.
.. . .
.. .. .
. . .
..
..
150LAL
10
..............
AA-
800
600
.
. . ...
. .
..
...
400
200
0
12AM, 01/24
12AM, 01/25
12AM, 01/26
12AM, 01/27
12AM, 01/28
12AM, 01/29
12AM, 01/30
12AM, 01/31
Time
Figure 5-4: Unit Commitment with Demand Response, 1/24-1/30
(2020, Natural Gas, Gradual Demand Growth, DR Price=50%, DR Participation=50%)
Despite January being an off-peak month, both basic forms of demand response included
in this model-curtailment and shifting-are evident. Curtailment service provided by loads
72
which require no recovery, such as lighting, elevators, and hotel signage, can be identified
in the figure as an increase in generation (i.e., a reduction in load) with no recovery period
before or after. These curtailments are visible in the mornings of January 25 and 27 and
the evenings of January 28 and 29. These four curtailments occur at four local maxima of
conventional generation due to, in chronological order: the onset of CSP generation prior
to PV's peak, the onset of PV generation on a day with poor early morning CSP potential,
and two daily peaks.
Load shift events can be identified in the figure as matching pairs of generation increases
and reductions; these correspond to a load curtailment followed by a recovery phase or, in
some cases, the reverse (such as when thermal loads are pre-cooled). These off-peak shifting
events are largely not correlated with reducing demand peaks, but rather with reduced
ramping of thermal generation to balance renewables. This behavior occurs on January
27, when each pair of load reductions and increases corresponds to a fluctuation in wind
generation. Its impact is clear on the active steam turbine units, which operate at a more
steady level. This is economical in the model because demand response is more expensive
per megawatt-hour than the steam turbines, but it has no ramping cost component.
Figure 5-6 illustrates a period in July, near the peak demand of the year, when high
temperatures and heavy usage of air conditioning drive demand. However, the daily peak
conventional generation is not significantly higher than in January, due to high penetration
of solar generation. Solar generation produces at a higher output level in summer (PV more
so than CSP), and also coincides better with peak cooling demand in the summer than it
does with peak heat demand in the winter. This has the effect of lowering and spreading
out peak demand, making the summer week shown in Figure 5-6 less conducive to peak
demand shaving than the winter week shown in 5-4.
All of the demand response events in this period are load shifts, seen by the increases
and reductions in demand in close proximity to one another. The least constrained load
shift in the model comes from residential dishwashers, the use of which can be postponed
for up to 12 hours; the large shift on July 23 is an example of a relatively long shift which is
not practical with thermal loads. Similar to the load shifts described in January, the shifts
shown in July largely act to balance renewable fluctuations and reduce ramping of other
units (seen by the dishwasher recovery period, which coincides with an increase in wind
production).
73
CCGT
ICE
ST
CSP+TES
CSP ---- -
PV
Wind
1000
.
800..
.
600
400
200
0
-150
C
a.)
a)
U
10 0
.. ..
. . . . . . .
.. . . . . . . . ..-
0
200
C
150
..
50
50
150
100
1000
-
50 0
0
12AM, 07/18
12AM, 07/19
12AM, 07/20
12AM, 07/21
12AM, 07/22
12AM, 07/23
12AM, 07/24
12AM, 07/25
Time
Figure 5-5: Unit Commitment in Baseline Case, 7/18-7/24
(2020, Natural Gas, Gradual Demand Growth, DR Price=oo, DR Participation=0%)
This time period also illustrates the use of demand response as a source of reserve. For
example, demand response can provide overnight reserve, allowing more steam units to be
cycled down (thus allowing the more efficient ICEs to remain more fully loaded). Also, a
drastic change in output from the CSP with storage is clear on July 19 and 20. The cause of
this change is not immediately obvious, but it is indirectly caused by the need for reserves
to balance wind intermittency, which can be explained in several steps:
1. In this time period, the three natural gas-fired Vasilikos steam turbines are providing
reserve. In the model, this constrains their ability to operate at maximum output,
since they must have the ability to ramp up generation.
2. This reserve constraint forces several of the older, heavy fuel oil-fired Dhekelia steam
74
m
CCGT
ICE;-
IST
CSP+TESE
CSP
- PV
Wind
DR
1000
800
600
400
200
0
0
-5
150
0
100
$-
50
0
200
150
100
50
0
I
A
AL'
LA &h A IL.....
..........
150
100
50
0
1000
500
0
12AM, 07/18
12AM, 07/19
12AM, 07/20
12AM, 07/21
12AM, 07/22
12AM, 07/23
12AM, 07/24
12AM, 07/25
Time
Figure 5-6: Unit Commitment with Demand Response, 7/18-7/24
(2020, Natural Gas, Gradual Demand Growth, DR Price=50%, DR Participation=50%)
units to briefly turn on each day.
3. In order to maintain balance between supply and demand, the CSP units with available
storage-which cannot provide spinning reserve-must ramp down and load their storage
instead.
4. The cost of CSP's storage decay and less efficient part-load operation is outweighed
in this case by startup costs, so the units output at a lower level until storage is
exhausted.
75
. . .......-
................
.
- ............ ....
The unit commitment with demand response in Figure 5-6 is more simple: demand response
provides the necessary reserve capacity instead of the heavy fuel oil-fired Dhekelia steam
units, and there is no need for the CSP generation to use its storage.
5.2.2
Operational Patterns of Demand Response
The unit commitment scheduling of demand response is analyzed further here in terms
of each type of demand response and its mode of operation during a load curtailment or
shift. The operating patterns of loads that are curtailed with no leading or lagging recovery
period are relatively simple. Such loads are curtailed in this model when they help avoid
startup or ramping up of conventional generation, whether to meet higher demand or to
balance drops in renewable generation.
However, load curtailment is only active in the
hours during which demand is high enough that the conventional generation being offset is
more expensive than curtailment, after factoring in the effects of ramping costs. Figure 5-7
illustrates the magnitude of curtailment in each hour of the year.
MW
10
1/1/2020
8
----- - - -
3/1/2020
6
4
5/1/2020 -
2
0
7/1/2020 [
Mo-
2
C/1/2020
-
11/1/2020
F
-4
-6
-8
12/31/2020 1L
00:00
06:00
12:00
Time of Day
18:00
24:00
-10
Figure 5-7: Impact of Curtailment on Demand by Hour
(2020, Natural Gas, Gradual Demand Growth, DR Price=50%, DR Participation= 50%)
76
---
-
- - -- -9900
LLLL
0 090LL
The operation of curtailment is most frequent in off-peak seasons, when the daily evening
demand peak is more pronounced relative to the daily low and more brief (meaning it is also
more amenable to being shaved for a short period of time). It is also used in certain months,
particularly January, when there is a separate morning peak in conventional generation,
largely due to the onset of demand prior to solar resource availability.
0
-1
-32
S -3
z
-5 Load Reductions
-61
0
1
1000
2000
3000
4000
5000
6000
7000
Hours, sorted by net demand level before DR
1
8000
Figure 5-8: Impact of Curtailment on Load Duration Curve
(2020, Natural Gas, Gradual Demand Growth, DR Price=50%, DR Participation=50%)
Figure 5-8 illustrates the impact of load reduction on the annual load duration curve.
The figure shows the hours of the year during which curtailments at some level occur; the
hours are sorted in descending order by net demand (i.e., after generation from renewables is
subtracted from electricity demand, but before the impact of demand resposne is factored
in).
No curtailments occur outside of high-demand hours.
The threshold at which no
curtailment occurs is at a net demand of roughly 425 megawatts, as evidenced by Figure
5-1. 432 megawatts is the total installed capacity of Cyprus's six ICE units and two CCGT
units in this scenario. 3 This suggests that curtailment is used almost exclusively to offset
generation or ramping from the steam units which handle most load following.
The operational role of load shifts is less simple than that of curtailments, as shown by
Figure 5-9. Like curtailment events, shift events are rarely used to reduce demand during
3
In this scenario, one open cycle gas turbine on one of the two CCGT units operates independently from
the CCGT configuration, as discussed in Section 4.2
77
4
0
-4
Load Increases
Reductions
-Load
-6
0
1000
6000
7000
2000
3000
4000
5000
Hours, sorted by net demand level before DR
8000
Figure 5-9: Impact of Load Shifts on Load Duration Curve
(2020, Natural Gas, Gradual Demand Growth, DR Price=50%, DR Participation=50%)
the hours when generation from CCGT and ICE units fully meets demand. However, it is
still used occasionally in those hours, both to increase demand and to reduce it. There is
no immediately obvious pattern in terms of which high-demand hours are targeted. Figure
5-10 shows the sum of increases and decreases from load shifts at each hour of the year,
sorted by date and time of day instead of by net demand level.
As in the case of curtailment, load shifts are used to reduce the daily demand peak on
some days when it is relatively pronounced (i.e., when evening heating loads replace midday
cooling loads). This is seen on days when a blue reduction in demand is preceded by a red
increase in demand-suggesting the use of a thermal load which can be shifted forward in
time, rather than an end use which could be delayed, such as a laundry machine. There
are also certain days of year-mostly in the winter, when the evening peak is pronounced-in
which demand response is used to simultaneously shift demand in both directions off a peak.
One reason the operational trends in Figure 5-10 are difficult to discern is that the effect
of balancing wind fluctuations, which can require shifts in either direction at any time of
day or year, adds noise to the effect of reducing peaks. Figure 5-11 disaggregates the effects
at each hour into discrete shift events. It then plots the cumulative effects of shift events
tied to a demand reduction starting at a given time of day. Depending on the type of end
use and its mode of operation in a given demand response event, the load reduction can
78
MW
1/1/2020
10
-
8
3/1/2020
6
4
5/1/2020
-[
2
7/1/2020
0
-2
9/1/2020
-4
-6
11/1/2020
-8
.
12/31/2020 1
00:00
.
06:00
12:00
Time of Day
.
.
18:00
24:00
-10
Figure 5-10: Impact of Load Shifts on Load Demand by Hour
(2020, Natural Gas, Gradual Demand Growth, DR Price=50%, DR Participation=50%)
MW
12AM
100
Day 1
X
A-
6AM
Day 1
N'
s
50
Delayed
Load
12PM
Appliance
Load Shifts ~
Day 1
0
-e
C1
Advan ced
Load
6PM
Day 1
U)
4D
-50
Thermal Load
Preconditioning
12AM
Day 2
12 PM
Da Ly 0
V-
6PM
Day 0
d
12AM
6AM
12PM
6PM
12AM
6AM
12PM
Day1
Day1
DayI
Day1
Day2
Day2
Day 2
Hours Impacted by Load Reduction at Time
-100
T
Figure 5-11: Cumulative Impact of Load Shifts Per Start Time of Load Reduction
(2020, Natural Gas, Gradual Demand Growth, DR Price=50%, DR Participation=50%)
79
either precede or follow the load reduction (or both).
One clear trend is that demand shift events are concentrated around evening peak hours,
where the annual cumulative load reduction is largest.
This seems correlated with the
operation of demand response in winter months shown in Figure 5-10. A much more subtle
concentration of events occurs from curtailments at 2 PM, which is the peak net demand
hour on certain summer days. One other trend is that in the hours preceding peak demand,
demand response is only used to shift demand forward in time. This is intuitive; it is rarely
economical to delay demand to occur at a peak hour.
300
200 100 -
0
-100 -200
-
-300 -400
-
Load Increases
Load Reductions
-500 -5
5
0
10
15
Hours Offset from Start of Load Reduction
Figure 5-12: Aggregated Cumulative Operating Mode of Load Shifts
(2020, Natural Gas, Gradual Demand Growth, DR Price=50%, DR Participation= 50%)
It also is evident at certain hours of the day that load shifts have two discrete varieties:
shifts forward in time that have the repeated effect of shifting load only one or two hours,
and delays which have a much more dispersed effect of increasing load anywhere from 1
to 12 hours later. Figure 5-12 highlights that trend by portraying the cumulative effects
at each hour as an offset from the start of load reduction, regardless of the time of day at
which the reduction occurred.
This effect is at least partially due to the modes of operation each end use in this
model; Figure 5-13 shows the modes of operation employed by each shiftable (as opposed to
curtailable) end use that is shifted at least once in this scenario. The modes of operational
80
Residential Refrigerators & Freezers
Residential Dishwashers
20
200
0
-20
-200
-40
.
-
Load Increases
Load Reductions
Load Increases
Load Reductions
-60
--400-
-5
10
5
0
Hours Offset from Start of Load Reduction
--5
15
Commercial HVAC
0
5
10
Hours Offset from Start of Load Reduction
15
Desalination
0.6
10
0.4
5
0.2
0
0
3
a5
-0.2
-
-10
Load Increases
Load Reductions
a -0.4
-0.6
-5
0
5
10
Hours Offset from Start of Load Reduction
Load Increases
Load Reductions
-15
15
-5
0
5
10
Hours Offset from Start of Load Reduction
15
Figure 5-13: Cumulative Operating Mode of Active Load Shift Programs
(2020, Natural Gas, Gradual Demand Growth, DR Price=50%, DR Participation=50%)
behavior shown here are constrained by the load characterizations described in more detail
in Chapter 3. For example, residential dishwashers are defined to have a load length of two
hours and can be shifted for up to 12 hours; this is why all load reductions last for exactly
two hours, and the shifts are widely dispersed in time. Desalination plants have flexible
load reduction lengths, and a moderately flexible shift lengths of up to four hours. The
further in time a load can be delayed or advanced, the less binding that constraint is, and
the more dispersed the cumulative effects of that end use will be when summed across the
year. It also allows for more dispersed effects for a given demand response event, suggesting
there is a value in aggregating loads which are more flexible and less constrained; recovery
effects could either be concentrated at another time or evenly dispersed, depending on what
is more optimal.
The binding nature of these constraints and the limited, "lumpy" selection of load types
create operational niches that certain loads fit into. For example, because consumers seem
willing to shift usage of dishwashers, their price to be shifted is the lowest and the shift length
81
is least constrained; furthermore, they have no loss factor to recover. This results in heavy
use of that option to delay loads, when it is optimal for the system. However, dishwashers
are constrained to only delay loads, and cannot increase consumption before reducing it.
Bringing consumption forward in time is handled in this model primarily by thermal loads.
As a rough approximation to model differences in thermal storage capabilities, refrigerators
and freezers are assigned a thermal loss factor of 5% per hour, while HVAC systems have a
loss factor of 10% per hour. When these thermal storage losses are factored in, refrigerators
and freezers are the lowest-price option to bring consumption forward in time, so they are
used heavily for those operations. However, HVAC systems are allowed to shift consumption
for a slightly longer period of time, and there is at least one instance in which that marginal
added flexibility is leveraged.
Including a greater diversity of load types in a demand
response program can allow for each load to be used only in the niche cases it is best suited
to.
5.2.3
Demand Response Value Assessment
One of the most visible impacts of demand response, as seen in Figure 5-4, is to react
to and balance fluctuations and forecast errors of wind power, thus allowing conventional
generation units to operate at a smoother output level. In this scenario, the use of demand
response reduced startup costs by 25.94%, although startup cost reduction accounts for less
than 20% of total cost reductions.
A larger impact of demand response is to provide an extra option for reserves. While
there are no days in this scenario with insufficient reserve margins, the low-cost reserve
provided by demand response allows other units to devote less time to operating reserve,
making their output levels less constrained and allowing them to operate nearer to full load
(or to be shut off in some cases).
Figure 5-14 shows the loading levels of conventional generation. The effect of balancing
renewables and that of providing reserve is apparent in the loading levels. Certain units that
are responsible for balancing of renewable fluctuations, such as the Vasilikos steam units
and, to a lesser extent, the ICE units at Dhekelia, spend much time at part load ramping
up and down. When demand response handles those fluctuations, the conventional units
are able to spend more time at their optimal output level, which is often either fully on
or off. The figure also shows that demand response removes the need for certain units to
82
Akrotiri CSP Trough
1
1
-Al'&
0.5
0.5
0
0
0
Pentakomo CSP DSW
-
1
0.4
0
Dhekelia ICE 1 1
0.5
Dhekeli
ICE 1
0
0
0.59
0
Dhekelia ICE
1
0
13
1
0.5
i 7M
0.5
0
0
0
1
0.59
0
1
0.5
0.5
0
0
0.59
1
Vasilikos CCGT 1
1
0.59
0
Vasilikos CCGT 2
1
1
0.5
0.5
0
0
0
0.55
1
Vasilikos Steam 1
Vasilikos Steam 2
1
0.5
0.5
0I
0.46
1
0.55
0
1
0
1
Dhekelia ICE 2 3
Dhekelia ICE 2 2
0
0.59
0
1
0
0.59
Dhekelia ICE 2 1
1
0
1
1
0.5
-e
0.4
Dhekelia ICE 1 2
1
0
V
-
0
0
~ri\V
I
0.46
1
Vasilikos Steam 3
Baseline
1
DR Price = 50%,
DR Participation
Min load
0.5
01
0
0.46
-
50%
1
Load Factor
Figure 5-14: Impact of Demand Response on Load Factors of Dispatched Units
(2020, Natural Gas, Gradual Demand Growth, DR Price=50%, DR Participation=50%)
83
operate at their minimum load factor at certain times when extra reserve is needed; this is
most pronounced on the ICE units.
To a certain extent, this effect is apparent on the CSP units with storage. The CSP units
with storage are only partially dispatchable, and the availability of direct solar irradiation
impacts the load factor at any given time. Although the CSP units do not actively provide
spinning reserve, there are still times at which they must operate at low load factors in order
to "make room" for conventional spinning reserve units to be activated to provide reserve,
as seen in Figure 5-5.
Units
Total Average Costs
DR Average Costs
DR Usage
S /MWh consumed
4 /MWh consumed
Average MWh/day
Natural Gas Usage
Fuel Oil Usage
CO 2
CSP+TES Capacity Factor
GJ
GJ
Average kg/MWh
%
Wind Curtailed
%
Baseline
DR Price=50%,
(No DR)
DR Participation=50%
58.67
-
57.93
0.02
0.85
2,198,069.74
1,822.75
351.77
41.20
2,174,963.30
72.64
347.72
41.95
0.04
0.00
Table 5.2: Summary of Demand Response Impacts
(2020, Natural Gas, Gradual Demand Growth, DR Price=50%, DR Participation= 50%)
Table 5.2 presents several metrics to assess the value of demand response in this scenario.
The average system costs are lowered by 460.72/MWh of total consumption (or 1.26%)
after factoring in S 0.02/MWh of total consumption of demand response energy payments.
Put another way, the annual avoided cost is F 3.62M after factoring in demand response
payments totaling S 0.11M. Most of the other results, including a 1.05% decrease in natural
gas usage and a near complete decrease in fuel oil usage, reflect the more optimal operation
of efficient, low-cost units and shutdown of inefficient, expensive units.
One interesting finding relates to the maximization of generation from installed renewables. Very little wind is curtailed in the base case, due to a low modeled capacity factor for
wind and sufficient (but relatively inefficient) reserves-only 0.123 GWh over the year, but
those occasional curtailments are fully eliminated. On the other hand, CSP with storage is
able to provide an additional 3.6 GWh of generation, boosting its capacity factor by 0.75%
to 41.95%.
This is due to its steam unit operating at a more efficient heat rate and its
storage being relied upon less heavily, reducing losses from hourly storage decay.
84
5.2.4
Impact of Fast Economic Growth
In a scenario with natural gas generation, if the economy of Cyprus rebounds from its
current economic crisis much more quickly than expected, the value of demand response is
not fundamentally changed. A set of cases is run assuming a 2020 annual peak demand of
1275 megawatts, which corresponds to a 31% increase in total annual energy consumption
from 2012; this matches the peak demand forecast in CERA's 2011 annual report (prior to
the explosion at Vasilikos and the economic crisis). The resulting load duration curve is
shown in Figure 5-15.
1400
Demand
Demand minus Wind
Demand minus Wind and PV
Demand minus Wind, PV, and CSP
1200
1000
800
0
600
0
400
200
0
SI
0
1000
2000
I
3000
4000
5000
Hours
I
I
6000
7000
8000
Figure 5-15: Load Duration Curve
(2020, Natural Gas, Fast Demand Growth, No Demand Response)
In the fast growth scenario, the model assumes a third CCGT system is built at Vasilikos
to meet demand. The model also scales up the capacity of each demand response option at
the same rate as total demand growth. Full results for these cases are given in Appendix C.1
and Table 5.3 summarizes the impacts of demand response.
Most of the changes from baseline shown here are larger in magnitude than in the case
with gradual demand growth, simply because total consmption is larger. Because demand
is scaled linearly, the portion of generation provided by baseload increases by the same
proportion as the gap between daily troughs and peaks. The increase in base demand is
85
Units
Total Average Costs
DR Average Costs
DR Usage
E /MWh consumed
E /MWh consumed
Average MWh/day
Natural Gas Usage
Fuel Oil Usage
CO 2
CSP+TES Capacity Factor
Wind Curtailed
GJ
GJ
Average kg/MWh
%
%
Baseline
DR Price=50%,
(No DR)
DR Participation=50%
61.60
-
60.83
0.02
1.08
3,013,371.84
5,424.33
369.22
40.94
0.00
2,985,413.35
637.09
364.98
42.17
0.00
Table 5.3: Summary of Demand Response Impacts
(2020, Natural Gas, Fast Demand Growth, DR Price=50%, DR Participation=50%)
met almost fully by the third CCGT unit (and the operating CCGT units operate more
steadily at full output, with less need to ramp down generation at certain hours).
This
explains the large increase in natural gas usage compared to the slow economic growth case.
The increase in the gap between daily troughs and peaks requires more daily cycling,
handled in this scenario by steam turbines. This is the the main difference with the scenario
including gradual economic growth. It explains why the capacity factor of CSP with storage
gets a slightly larger boost in this scenario: when more daily ramping up and down of steam
turbines (with relatively high startup costs) is needed, it is more optimal to use storage to
shift generation to the peak demand hours.
5.2.5
Sensitivity to Demand Response Price and Participation
Two key parameters in the design of an event-based demand response program are the program's participation level-determined at least in part by the payment given to participants
for enrolling-and the level of payments given to reimburse participants per megawatt-hour of
load that is curtailed or shifted. Because those parameters are not optimized by WILMAR,
but rather are input parameters, several price and participation levels were used to test how
sensitive the model is to changes.
One useful metric for price and participation is simply how heavily demand response is
utilized over the course of the year. Figure 5-16 shows the average daily energy output of
demand response programs designed with various energy payment prices per megwatt-hour
and participation levels, normalized such that a value of 1 is equivalent to 0.85 MWh/day,
the value obtained when the price and participation levels are each 50%.
86
9.0 -
8.20
8.0 -
-7.0
3
-
6.0 -
Q 5.0 .0 -
30% Participation
6 4.
3.0 -
50% Participation
a70% Participation
2.0 1.0 -
1.00
0.61
0.73
0.00.03
30%
50%
70%
DR Price
Figure 5-16: Normalized Average Daily Demand Response Usage
(2020, Natural Gas, Gradual Demand Growth, Stochastic Simulations)
Lowering the price level significantly increases demand response utilization.
At 30%
participation levels and prices at 70% the cost of a peaking natural gas turbine, demand
response is almost never used; if prices are lowered to 30%, then it is used 73% as often as
the 50%-50% case. Similarly, at 70
The usage of demand response is equally sensitive to the program's participation level.
At each price point, increasing the participation level from 30% to 70% results in a greater
than tenfold increase in utilization.
The program capacity is not necessarily a binding
constraint in any modeled scenarios, and the usage is not necessarily expected to increase
linearly with participation. One factor at play is that certain end uses may be too tightly
constrained, or too expensive, to be regularly curtailed or shifted regardless of their sizing;
those end uses may still be useful as a form of reserve.
Another useful metric to compare the impacts of demand response across scenarios is
the average load factor of all dispatchable units, weighted by the capacity of each unit. This
metric is used in [23], where it is referred to as an "efficiency score". The efficiency score's
value is 0 if the average load factor equals the minimum load factor, and 1 if the average
load factor equals 1 (full output). Figure 5-17 shows how demand response program design
impacts the efficiency score, normalized such that a value of 1 is equivalent to an efficiency
score improvement of 28.2%.
Efficiency is impacted by program design in a very different way than demand response
utilization; lowering the pricing has almost no effect on efficiency, despite changing how
often demand response is utilized. This may be explained by the role of operating reserves.
87
1.4
1.2
1.0
0.8
M 30% Participation
q
0.6
~
50% Participation
m70% Participation
.4 -
0.4
o
4
0.2
-
0.0
30%
50%
70%
DR Price
Figure 5-17: Normalized Increase in Efficiency Score, Weighted by Generator Capacity
(2020, Natural Gas, Gradual Demand Growth, Stochastic Simulations)
Regardless of pricing or operational constraints, the provided demand response end uses
increase the set of available options for providing spinning or non-spinning reserve, and the
need to provide reserves is why many conventional generators are online at inefficient output
levels in the baseline case.
Figure 5-18 shows how program design impacts the reduction in emissions of CO 2 ,
normalized such that a value of 1 is equivalent to an emissions reduction of 1.2%. While no
carbon tax is included in the model (although WILMAR is capable of including one), it is
still a relevant metric in for the value of demand response.
1.2 -
~1.0
V
>
-
0.8
)
0
M30% Participation
0.6 -
M 50% Participation
U0
70% Participation
.
0.2 II
0.0
30%
50%
70%
DR Price
Figure 5-18: Normalized CO 2 Emissions Reduction
(2020, Natural Gas, Gradual Demand Growth, Stochastic Simulations)
Program design's impact on emissions is similar to the impact it has on efficiency, due
to the strong correlation between those two metrics. Emissions are driven entirely by fuel
88
consumption, which depends on generator efficiency (WILMAR models extra costs due to
generator ramping, but does not explicitly include extra fuel consumption). This means
the emissions reduction, like efficiency improvements, depends on participation levels more
than price.
89
5.3
Results for 2020 with Fuel Oil Generation and Gradual
Demand Growth
5.3.1
Unit Commitment Scheduling
It is worth considering whether responsive loads are valuable in Cyprus's future unit commitment scheme if the planned switchover to natural gas electricity generation does not
occur. Given that the switchover is several years away and many questions remain to be
answered regarding the usage of new gas discoveries, it is conceivable that the switchover
may not occur. This scenario is identical to the one discussed in Section 5.2, except that
no generators are converted to natural gas. Instead, all steam and internal combustion
engine units burn heavy fuel oil, while all gas turbines burn #2 diesel oil. The resulting
load duration curve is shown in Figure 5-19.
1000
Demand
Demand minus Wind
Demand minus Wind and PV
900
800
_____Demand
minus Wind, PV, and CSP
-
700
600
0
500
400
0
U)
300
200
100
01
0
1000
I
2000
I
3000
I
1
4000
5000
Hours
6000
7000
I
8000
Figure 5-19: Load Duration Curve
(2020, Fuel Oil, Gradual Demand Growth, No Demand Response)
As shown in Figure 5-20, relying on fuel oil for generation in 2020 changes the ranking
of units by operational costs. Combined cycle units at Vasilikos use #2 diesel oil, which is
more expensive than the heavy fuel oil used by the steam and internal combustion units.
Those heavy fuel oil-burning units become the new baseload generation, while the CCGT
90
SI
Larnaca CSP Stirling -50.8 MW
Nicosia CSP Stirling -25.5 MW
Solar PV -192.0 MW
Wind -300.0 MW
Pentakomo CSP DSW 4.5 MW
Akrotiri CSP Trough 50.0 MW
DR Res. Dishwasher
DR Res. Washer-Dryer
DR Res. Washer
DR Desal
Dhekelia ICE 2-3
Dhekelia ICE 2-2
Dhekelia ICE 2-1
Dhekelia ICE 1-3
Dhekelia ICE 1-2
Dhekelia ICE 1-1
DR Comm. HVAC
DR Res. HVAC
Vasilikos Steam 3
Vasilikos Steam 2
Vasilikos Steam 1
DR Res. Refrig-Freezer
Vasilikos CCGT 2
Vasilikos CCGT 1
DR Comm. Curtailments
Dhekelia Steam 6
Dhekelia Steam 5
Dhekelia Steam 4
Dhekelia Steam 3
Dhekelia Steam 2
Vasilikos CCGT 1 Peak
Vasilikos OCGT 1
Moni OCGT 4
Moni OCGT 3
Moni OCGT 2
Moni OCGT 1
0
-I
I
Non-fuel variable costs
Fuel cost at maximum load
Extra fuel cost at minimum load
100
200
300
400
Variable costs (euros/MWh)
500
600
700
Figure 5-20: Variable Costs Per Unit
(2020, Fuel Oil, Gradual Demand Growth, DR Price=50%, DR Participation=50%)
91
units are at the middle of the merit order.
I
ICE
ST
CCGT
CSP+TES
CSP [
_
PV
Wind
1000
800
-
600
400
200
0
200
-.
150
.
.
.. . . . . ..
100
.. .
. ... . . .
50
0
Q)
..... .. .. . ................... ... .
150
-
100
50
0
150
1 00
5
. . . .. .
0
..
..-.
.. .
..
. ..
.. . . . . . . .. ..
. . . . ... .
04
800
6 00
.. .
.. . . . .
.. . . .
.. . .
. ..
.....
400
200
0
12AM, 01/24
12AM, 01/25
12AM, 01/26
12AM, 01/27
12AM, 01/28
12AM, 01/29
12AM, 01/30
12AM, 01/31
Time
Figure 5-21: Unit Commitment in Baseline Case, 1/24-1/30
(2020, Fuel Oil, Gradual Demand Growth, DR Price=oo, DR Participation=0%)
Figure 5-21 illustrates the daily cycling on and off of CCGT units to meet evening peak
demand in January, as well as morning peaks on days when demand rises earlier than solar
output.
Another effect of reliance on fuel oil for generation is that the large variable cost gap
between certain units due to different fuel sources does not occur. This enhances the value
of demand response, which is priced according to the cost of a peaking gas turbine unit.
In this case and the case of natural gas generation, the peaking units are fired using #2
diesel oil, so their absolute costs (and the absolute costs of demand response) are constant
in all modeled scenarios. However, without natural gas, the other units are not drastically
cheaper, making a much more frequent usage of demand response economical-in fact, de-
92
ICE
ST
CCGT
Peakerm
CSP+TES
CSP I
-
PV
Wind
-
DR
1000
800
-.
.
...
.
600
400
200
0
10
5
200
S150
100.
50
010--100
150
0
850
150
400
12AM, 01/24
12AM, 01/25
12AM, 01/26
12AM, 01/27
12AM, 01/28
12AM, 01/29
12AM, 01/30
12AM, 01/31
Time
Figure 5-22: Unit Commitment with Demand Response, 1/24-1/30
(2020, Fuel Oil, Gradual Demand Growth, DR Price=50%, DR Participation= 50%)
salination and shiftable residential appliances are cheaper to dispatch than any conventional
units.
The effects of this change in merit order are evident when comparing Figure 5-22 to
Figure 5-4, which reflects an identical optimization problem aside from its inclusion of
natural gas. With no natural gas, demand response is called in all of the same instances
as with natural gas, plus various other instances. It is called at nearly every day's evening
peak, nearly every morning net demand peak, and is called at least once (in the evening of
1/25) to mitigate generator ramping due to a sudden increase in wind production.
Compared to the case with natural gas, CSP with storage no longer stores energy into
93
ICE
ST
CCGT
CSP+TES
CSP
[~~~~jPv
Wind
1000
t~1~f-A
800
600
400
200
0
150
C
100
50
0
- -
200
150
100
50
-/
L
\\
.
/-.A
-
...
......
/-
0
150
100
50
0
1000
500
0
12AM, 07/18
12AM, 07/19
12AM, 07/20
12AM, 07/21
12AM, 07/22
12AM, 07/23
12AM, 07/24
12AM, 07/25
Time
Figure 5-23: Unit Commitment in Baseline Case, 7/18-7/24
(2020, Fuel Oil, Gradual Demand Growth, DR Price=oo, DR Participation=0%)
the late evening to ramp back up and help meet evening peak demand in this particular
week. This is partly because demand response is a cheaper option. It is also partly because
CCGT units, no longer operating at full output constantly due to high fuel oil costs, provide
an option to ramp up and down quickly with relatively low startup costs. When natural
gas is available, steam generators become the load-following units which must ramp up and
down, and their startup costs are twice as high as CCGT generators.
Figures 5-23 and 5-24 show similar results. When demand response is available at low
cost, it is used frequently for a wide range of uses, including reduced generator ramping,
reduced overnight turndown of steam units, assistance with filling in gaps in renewable
output (such as the lull in wind output around 12AM on 7/21), and avoided usage of CSP's
storage capabilities.
94
ICE
ST
CCGT
CSP+TES
]
CSP
PV
Wind
DR
1000
800
/*-" . ...
../-Ik
600
200
0
2
0
-2
-4
150
C
C
C
0
100
. .
".
............
50
0
L
200
0
150
100
50
0
150
.
.
.
.
..........
.. .
....
....
.
.
..
... ...
AA AA AZL
\ . . .. .
. ..
12AM, 07/19
12AM, 07/20
. . .. .
. .. .
. . ..
100
50
0
1000
500
0
12AM, 07/18
12AM, 07/21
12AM, 07/22
12AM, 07/23
12AM, 07/24
12AM, 07/25
Time
Figure 5-24: Unit Commitment with Demand Response, 7/18-7/24
(2020, Fuel Oil, Gradual Demand Growth, DR Price=50%, DR Participation=50%)
5.3.2
Operational Patterns of Demand Response
Curtailment and shifting of loads play a more integral role in the daily unit commitment
schedule when they are less expensive relative to other units. At first glance, this makes it
difficult to find specific patterns shared by the cases in which demand response is used.
Figures 5-25 and 5-26 do not show many obvious trends guiding the usage of demand
response. This is a contrast to the case in which natural gas generation is used. In that
case, curtailment has little role to play during the hours of the year when only relatively
95
1W1
-1
-2
-3
S-4
Z -5
-6
Load Reductions
0
1000
2000
3000
4000
5000
6000
7000
Hours, sorted by net demand level before DR
8000
Figure 5-25: Impact of Curtailment on Load Duration Curve
(2020, Fuel Oil, Gradual Demand Growth, DR Price=50%, DR Participation=50%)
6
-
4 1
W
I
a .
20
0
-2
z
t-4
I
-6 -80
-----
1000
Load Increases
Load Reductions
2000
3000
4000
5000
6000
7000
Hours, sorted by net demand level before DR
8000
Figure 5-26: Impact of Load Shifts on Load Duration Curve
(2020, Fuel Oil, Gradual Demand Growth, DR Price=50%, DR Participation=50%)
96
economical natural gas-fired baseload units are active, and shifts mostly move demand onto
those hours rather than away from them. However, when fuel oil is used for all generation,
there is no price gap discouraging demand response's use in low-demand hours.
MW
1/1/2020
10
8
3/1/2020
6
4
5/1/2020
2
7/1/2020
0
--2
9/1/2020
-4
11/1/2020
-6
1
-8
12/31/2020 1'.
00:00
06:00
12:00
Time of Day
18:00
24:00
-10
Figure 5-27: Impact of Curtailment on Demand by Hour
(2020, Fuel Oil, Gradual Demand Growth, DR Price=50%, DR Participation=50%)
Figure 5-27 helps illustrate several common cases in which commercial curtailment is
used throughout the year:
" The daily peaks in net demand which occur in the late evening, when solar production
is less active. This is the main case in which curtailment occurs in cases with natural
gas generation.
" The daily secondary peaks in net demand occurring briefly in the morning before solar
production ramps up (mostly in the winter and spring months when there are fewer
hours of full daylight).
" Early morning hours in the hottest weeks of the summer, when demand is still relatively high.
Demand response is called to allow the relatively expensive CCGT
97
....
. .........
....
. . ...............
...
production to be ramped down more quickly, thus reducing the amount of time it
operates at relatively low part-load efficiency. The daily low net demand in those
weeks is around 4AM, and CCGT units are ramped down between 12AM and 4AM.
MW
1/1/2020
.
10
~-
8
3/1/2020
6
4
5/1/2020 -
[
--
a)
7/1/2020
2
0
-2
9/1/2020
-4
-6
11/1/2020
-8
12/31/2020 L
00:00
06:00
12:00
Time of Day
18:00
24:00
-10
Figure 5-28: Impact of Load Shifts on Load Demand by Hour
(2020, Fuel Oil, Gradual Demand Growth, DR Price=50%, DR Participation=50%)
Figure 5-27 shows the usage of load shifts for each hour of the year. The usage is more
complex, since each load reduction must have a corresponding load increase at other hours,
but the usage is generally similar to curtailment. Load shifting is used on daily peaks in net
demand, daily secondary peaks in the morning when demand outpaces solar production,
and early morning hours in July and August (to help reduce output of CCGT units more
quickly). It shows one additional usage: from roughly 1PM-3PM throughout the year, it
is used on the early "shoulder" of daily peak demand, to delay the ramping up of CCGT
units. Like allowing earlier ramping down of CCGT units on the late "shoulder" of peak
demand, this does not affect ramping costs-the same amount of capacity is being brought
online or offine-but it does reduce the hours at which the units are operating at inefficient
partial load factors.
98
MW
Dayl
2M
L100
100
t6AM
4'
Day 1
A
Delayed
DLoad
Appliance
Load Shifts
12PM
_
Day 1
0
Advanced
Load
6PM
Day 1
4u'J
$_4
12AM
Day 2
50
-50
Thermal Load
Preconditioning
1
_..
12PM
6PM
12AM
6AM
Day0
12PM
Day0
6PM
Dayl
12AM
DayI
6AM
Day1
12PM
Day1
Day2
Day2
Day2
Hours Impacted by Load Reduction at Time
-0
-100
r
Figure 5-29: Cumulative Impact of Load Shifts Per Start Time of Load Reduction
(2020, Fuel Oil, Gradual Demand Growth, DR Price=50%, DR Participation=50%)
The usage of load shifting on the "shoulders" of daily cycling of CCGT units is also visible
in Figure 5-29. The heaviest usage of load shifting, seen by deep blue colors, is between
around 1-3PM and around 9-11PM. Appliances are utilized to shift midday loads around
1-3PM to the later demand "shoulder", as late as 3AM of the following day. (Appliance
consumption can only be delayed in this model, not advanced forward in time; only thermal
loads can be advanced in time). Shifts are also used to shift load in each direction from
the late evening peaks at 9-11PM. Thermal pre-conditioning is used much more heavily in
the evening hours than in the early afternoon, most likely because peak solar output can
be used to supply the advanced thermal load consumption in those hours.
Figures 5-30 and 5-31 give the cumulative usage of demand response load shifts, shown
in terms of load reductions and increases offset from the start of a load reduction phase. In
comparison to the similar output from the case with natural gas, seen in Figures 5-12 and
5-13, load shifts are used more than twice as heavily.
The same four end uses are utilized in this case as in the case with no natural gas,
despite overall demand response utilization being higher. This is explained by the fact
that the demand response capacity constraints are not truly binding, and the end uses that
are not utilized are strictly "dominated" by the end uses that are utilized. For
example,
residential HVAC is considered more inefficient than commercial HVAC (due to an assumption that the typical hotel is better insulated than the typical residence), so that option is
99
I
600
400 [
200 0
-200
-400 -600 -800 -1000-
Load Increases
Load Reductions
1200 14001
-5
0
5
10
15
Hours Offset from Start of Load Reduction
Figure 5-30: Aggregated Cumulative Operating Mode of Load Shifts
(2020, Fuel Oil, Gradual Demand Growth, DR Price=50%, DR Participation=50%)
Residential Refrigerators & Freezers
Residential Dishwashers
100
500
0
0
-100
500
-200
Load Increases
Load Reductions
-5
0
5
10
Hours Offset from Start of Load Reduction
'
15
-5
Commercial HVAC
Load Increases
Load Reductions
0
5
10
Hours Offset from Start of Load Reduction
15
Desalination
10
50
0
0
-10
-50
-
-100
Load Increases
Load Reductions
-20
-150
-5
0
5
10
Hours Offset from Start of Load Reduction
15
-5
Load Increases
LLoad Reductions
0
5
10
Hours Offset from Start of Load Reduction
Figure 5-31: Cumulative Operating Mode of Active Load Shift Programs
(2020, Fuel Oil, Gradual Demand Growth, DR Price=50%, DR Participation=50%)
100
15
never called upon to provide load shifting services. However, those less efficient or more
tightly constrained end uses can still provide reserve capacity, which explains the benefits
to increasing program capacity seen in Sections 5.2.5 and 5.3.5.
5.3.3
Demand Response Value Assessment
The fact that demand response is utilized more heavily without a natural gas switchover
suggests that its economic value is similarly increased. In fact, the modeled demand response
program would save the Cyprus electricity sector
versus
4
4
7.31M if the switchover does not occur,
3.62M if it does.
Much of the economic value of demand response arises from more optimal operation of
conventional generating units, and the change in average load factors shown in Figure 5-32
illustrates why. The Dhekelia internal combustion engines, the system's most economical
units, are able to operate at full output for nearly the entire year, rather than reducing
output to provide upward spinning reserve or to accommodate high amounts of wind when
demand is low. The Vasilikos steam generators, mid-merit units in terms of variable costs,
are able to spend less time at their minimum load factor and operate more like baseload
units instead. The Vasilikos CCGT units, efficient but relatively expensive due to the use of
#2 diesel oil, are left offline for more hours, and operate closer to full capacity when online.
The impact on CSP output levels is slightly more pronounced than with natural gas
generation (shown in Figure 5-14). In both cases, demand response allows CSP to spend
nearly 50% the year operating at full output and nearly 40% of the year shut off completely,
thus maximizing the efficiency of harvesting electricity from the sun.
However, in the
baseline case without natural gas, the CSP with storage units operate at minimum load
factor for a greater number of hours, so the increase from baseline is more pronounced. 4
Table 5.4 provides high-level metrics of the value of demand response when no natural
gas switchover occurs.
The average system costs are lowered by
consumption (or 1.77%) after factoring in
response energy payments.
4
4
1.72/MWh of total
0.09/MWh of total consumption of demand
Put another way, the annual avoided cost is 4 7.31M after
factoring in demand response payments totaling E 0.39M.
4
Without natural gas, CSP with storage spends more time operating at minimum output in order to
"make room" for other units to provide operating reserve. However, the baseline capacity factors are nearly
equal. This is likely because when all units use fuel oil, there is a smaller price gap between baseload units
and load following units, and less need for CSP to incur storage losses by shifting output to peak demand.
In the baseline cases, these two effects cancel out, making the capacity factors similar.
101
Akrotiri CSP Trough
Pentakomo CSP DSW
1
1
0.5
0.5
0
0
0
Dhekelia ICE 11
1
1
0.5
0.5
0
0
0.59
0
r
1
0.5
0.5
0.59
0
1
0.59
0
Dhekelia ICE 2 2
0.59
1
Dhekelia ICE 2 3
1
1
0.5
0.5
0
0
0.59
0
1
0
Vasilikos CCGT 1
1
1
'I IV\j N
0.5
0
0
0
0.55
1
0
Vasilikos Steam 1
r
0
1
0.59
Vasilikos CCGT 2
0.5
0
Dhekelia ICE 1 2
0
0
0.5
1
Dhekelia ICE 2 1
1
1
0.4
0
1
Dhekelia ICE 1 3
-o
a
p
0
1
0.4
7
-
0.55
---1
Vasilikos Steam 2
1
'I \V
0.5
0
0.46
1
0
0.46
1
Vasilikos Steam 3
1
Baseline
DR Price = 50%,
DR Participation = 50%
0.5
- Min load
0
0
0.46
1
Load Factor
Figure 5-32: Impact of Demand Response on Load Factors of Dispatched Units
(2020, Fuel Oil, Gradual Demand Growth, DR Price=50%, DR Participation=50%)
102
Baseline
(No DR)
97.22
-
DR Price=50%,
DR Participation=50%
95.50
0.09
3.91
2,401,342.87
530.04
2,374,399.99
523.57
Total Average Costs
DR Average Costs
DR Usage
Units
E /MWh consumed
4 /MWh consumed
Average MWh/day
Fuel Oil Usage
CO 2
GJ
Average kg/MWh
CSP+TES Capacity Factor
%
42.18
42.33
Wind Curtailed
%
0.01
0.00
Table 5.4: Summary of Demand Response Impacts
(2020, Fuel Oil, Gradual Demand Growth, DR Price=50%, DR Participation=50%)
Compared to the scenario with natural gas, this scenario sees a 353% increase in demand
response energy payments, versus only a 201% decrease in total operating costs.
This
could suggest that the most obvious usages of demand response-the "low-hanging fruit"are exhausted at some point, and there are eventually diminishing returns even when it
is still economical to use more demand response. Nonetheless, because baseline costs per
megawatt-hour are 62% higher when no natural gas is available, the magnitude of benefits
minus costs is still significantly higher without natural gas.
As in all modeled scenarios, there is relatively little curtailment of wind to be reduced
by demand response, due to low estimated capacity factors for wind. Demand response
increases the realized capacity factor of CSP with storage more drastically when there is
no natural gas. This is because there is less price gap between baseload and load-following
units, meaning less value in the use of storage to shift generation to evening peak demand
hours, so storage losses are lower.
5.3.4
Impact of Fast Economic Growth
As in the case of a natural gas switchover described in Section 5.2.4, if the economy of Cyprus
grows at a faster rate than forecast, the value of demand response is not fundamentally
altered. The demand is scaled in a manner identical to the case including natural gas; the
resulting load duration curve is shown in Figure 5-33.
Full results for this scenario are given in Appendix C.2, and Table 5.5 summarizes the
impacts of demand response. As in the scenario with a switchover to natural gas, most of
the changes in the metrics shown in Table 5.5 are scaled up with total generation levels. One
difference between the natural gas and fuel oil cases is that in this case, the capacity factor
103
1400
Demand
Demand minus Wind
1200
_
Demand minus Wind and PV
Demand minus Wind, PV, and CSP
1000
800
S
600
0
400
200
0
0
1000
2000
I
3000
I
I
4000
5000
Hours
1
6000
7000
1
8000
Figure 5-33: Load Duration Curve
(2020, Fuel Oil, Fast Demand Growth, No Demand Response)
Units
Baseline
DR Price=50%,
(No DR)
DR Participation=50%
Total Average Costs
DR Average Costs
S /MWh consumed
4S /MWh consumed
106.49
-
104.28
0.09
DR Usage
Fuel Oil Usage
CO 2
CSP+TES Capacity Factor
Wind Curtailed
Average MWh/day
GJ
Average kg/MWh
%
%
3,304,775.96
555.64
41.00
0.00
6.23
3,259,663.14
547.43
42.32
0.00
Table 5.5: Summary of Demand Response Impacts
(2020, Fuel Oil, Fast Demand Growth, DR Price=50%, DR Participation=50%)
of CSP with storage is relatively unaffected by demand growth. This is because the three
CCGT units handle most daily cycling instead of the steam units used for load-following
when there is natural gas. CCGT units have a startup cost of 6 20 versus 4 40 for the
steam turbines, and are much more efficient, so there is less incentive for the use of storage
to shift generation to peak demand hours.
In the case of higher demand growth, the relative amount of intermittent generation is
smaller than with gradual demand growth, lessening the integration issues involved. This
means that there is no wind curtailment even in the baseline case. If estimates of wind
104
production assumed a higher capacity factor, then the generation system would likely show
various signs of stress not seen here, including more wind curtailment, reserve requirement
shortages, and more usage of demand response to absorb intermittency and maintain grid
balance.
5.3.5
Sensitivity to Demand Response Price and Participation
Sensitivity analyses for demand response program pricing and participation levels are conducted using stochastic cases for fuel oil generation and gradual demand growth, following
the same approach as with scenarios including natural gas generation. Three of the most
relevant metrics for assessing sensitivity to pricing and participation are the usage levels
of demand response, the increase in conventional generator efficiency from the status quo,
and the reduction in carbon emissions. Figure 5-34 shows the average daily energy output of demand response programs, normalized such that a value of 1 is equivalent to 3.91
MWh/day, the value obtained in stochastic simulations when the price and participation
levels are each 50%.
3.0 2.55
2.5
VQ 2.0
1.5 -
*30% Participation
1.12
1.00
0 50% Participation
1.0 -M
70% Participation
0.5
0.05
0.12
0.0
30%
50%
70%
DR Price
Figure 5-34: Normalized Average Daily Demand Response Usage
(2020, Fuel Oil, Gradual Demand Growth, Stochastic Simulations)
Compared to the cases with natural gas generation, when fuel oil is used for all units,
demand response's usage is roughly equally dependent on price-increasing roughly by a
factor of 20 when price levels are lowered from 70% to 30%, but at first glance, is less
dependent on participation. Usage levels very roughly increase linearly with participation
levels, rather than the tenfold increase seen as in the stochastic simulations with natural
gas. This is likely explained by the fact absence of a large price gap between open cycle
105
peaking gas turbines (from which the price of DR is scaled) and less expensive natural gasfired units. Without that price gap, it becomes economical to call demand response much
more frequently, and the participation level is more likely to become a binding constraint.
Figure 5-35 shows how program design impacts the efficiency score improvement, normalized such that a value of 1 is equivalent to an efficiency score increase of 41.2%.
1.4 1.2 -
00
.
r 0.8 0 30% Participation
T
V0
Q)
E
M
0.6
* 50% Participation
m
.4
70% Participation
00.
04
0.2
0
0.0
50%
30%
70%
DR Price
Figure 5-35: Normalized Increase in Efficiency Score
(2020, Fuel Oil, Gradual Demand Growth, Stochastic Simulations)
The sensitivity of efficiency scores to program design parameters is similar regardless of
whether natural gas or fuel oil dominates the generation mix. Efficiency gains are relatively
sensitive to participation levels, but not to price levels. This is explained by the fact that
demand response need not actually be called to generate or consume electricity from the
grid, so price is less relevant. It is also noteworthy that at a high participation level, there is
relatively little to be gained by decreasing prices from 70% to 30%. This may be explained
by the fact that efficiency gains can only go so high: some level of inefficient part-load
generator output is necessary simply because the demand levels are continuous, and not
equal to the sum of generators' capacities.
Figure 5-36 shows how program design impacts the reduction in emissions of C0 2 ,
normalized such that a value of 1 is equivalent to an emissions reduction of 1.22%.
The sensitivity of emissions reductions to price and participation is similar to that of
efficiency, albeit with somewhat less variation for different prices and participation levels.
This trend is also seen in the scenarios including natural gas. This is because emissions
reductions are largely dependent on conventional generator efficiencies, which are relatively
106
1.2
-
1.0
-
0.8
-
0.6
~m
0
a
o
0
0 0.4SN
-
30% Participation
C450%
Participation
M 70% Participation
0.24 0.0
30%
50%
70%
DR Price
Figure 5-36: Normalized CO 2 Emissions Reduction
(2020, Fuel Oil, Gradual Demand Growth, Stochastic Simulations)
sensitive to program participation levels.
107
108
Chapter 6
Policy Implications
The results discussed in Chapter 5 suggest that demand response can provide a cost-effective
option to help ease Cyprus's transition to higher penetration of renewables. This chapter
discusses some limitations of the results, and synthesizes the results into a set of policyrelevant insights for Cyprus.
6.1
Limitations of Modeling Results
By necessity, the modeling approach and input data described in Chapters 3 and 4 create a
simplified representation of the Cyprus power system. Accordingly, the insights of the model
are useful in general terms, but several limitations should be noted. Certain limitations in
the model may lead it to overestimate the value of responsive demand relative to the results
presented here:
Perfect reliability of demand response participants: In this model, requests
from the demand response program operator are assumed to be honored
perfectly by the end user. This behavior is valid if the end uses are all
equipped with direct load control technology, but less valid if consumer intervention is required. This overestimates the dispatchability of end uses,
particularly shiftable appliances such as dishwashers, which may be more
difficult to fully automate. The impact of this assumption is somewhat
tempered by the counterbalancing assumption of perfect reliability of conventional generation.
109
No temporal component to demand response program capacity: The model
assumes that a given end use is equally available to be shifted or curtailed
at any hour of the day. In reality, certain loads are more or less useful at
certain times of the year or the day. For example, HVAC is used less heavily in the spring and fall than the summer and winter, and the more overall
consumption that occurs on a given day, the more consumption there is to
be modified.
However, several other limitations may lead to an underestimation of the value of responsive
demand:
Lack of a carbon price: The model finds that demand response tends to lower
carbon emissions purely as a side effect of other impacts, despite there being no modeled cost component related to emissions. This is due to higher
levels of conventional generation efficiency and greater usable output from
CSP and wind, as well as a marginal impact due to the curtailment of
certain loads which do not require a load shift and recovery. Carbon pricing was not included in this model, although WILMAR supports a tax on
emissions. If, in the future, EU carbon permit prices are sufficiently high to
modify the dispatch in Cyprus (or if an alternate policy scheme such as a
carbon tax is in place), then the value of demand response would increase,
particularly as an option to provide reserve without burning fuel, since the
program size and reserve capacity is what increases generator efficiency.
Perfect solar forecasting: Both direct normal irradiance and global irradiance are intermittent but perfectly predictable in the model. This is not
entirely unreasonable, since solar is more predictable than wind generation.
If solar forecast uncertainty would be included, there would be a greater
need for reserves in Cyprus in the daytime hours; one way of meeting those
reserves would be with demand response.
Low estimated wind generation: The scaled wind data has a relatively low
modeled capacity factor of 12.8%. Such a low value tends to reduce the
need for demand response as an active balancing mechanism (i.e., by calling
on customers to shift or curtail load).
110
No transmission and distribution costs: Due to a lack of available data,
the model does not include transmission and distribution losses, congestion,
or costs. Because end uses are geographically distributed than some other
options for grid stability, such as pumped storage, the model does not fully
reflect the cost impacts of demand response. This is particularly the case
if large-scale installation of distributed rooftop PV occurs, which would be
similarly scattered geographically and located near end uses.
It is difficult to predict the ways in which these assumption changes the overall high-level
impacts of demand response, and sensitivity to the assumptions warrants further investigation.
6.2
Program Design Considerations
The details of a demand response program's design are key to its success. While this model
does not optimize over different demand response program price or participation levels, the
sensitivity to program design parameters discussed in Sections 5.2.5 and 5.3.5 lend some
insights, seven of which are discussed here.
1. Demand response usage depends heavily on conventional generation costs.
A comparison of Sections 5.2 and 5.3 makes it clear that conventional generation costs
impact how heavily and frequently demand response is used. In the specific context
of the model, the impacts arise from the price gap between fuels (and, accordingly,
between fuel oil-burning peaking units and natural gas-burning baseload generation).
This is because in fuel oil scenarios, certain demand response options becomes cheaper
than every conventional unit, resulting in a fivefold increase in the dispatch of demand
response in the unit commitment schedule. In a more realistic situation, the impacts
due to various other constraints and cost components may have similarly strong influences.
2. Efficiency and emissions impacts depend mostly on participation levels.
In scenarios with and without a natural gas switchover, conventional generator efficiency is impacted much more strongly by program participation than pricing. This
is due to the value of reserve capacity provided by demand response, which can be
111
instantly summoned without burning fuels in "standby" mode. The same dependency
on program participation levels applies carbon emissions reductions, which are tied
closely to efficiency gains.
3. Active balancing of intermittency depends on both price and participation.
Actively balancing fluctuations in renewable generation (and the resulting startup cost
reduction) requires actively shifting or curtailing loads, rather than merely reducing
the need for generators to operate on standby. The resulting impacts are dependent
both on program size and participation levels.
4. High up front enrollment payments are preferable to high energy payments.
Up front enrollment incentives are not optimized within the model, but are assumed
to be sufficiently high to yield the modeled participation levels. Higher up front enrollment incentives could allow the program designer to increase program capacity and
lower energy payments while maintaining the same level of welfare for each participant.
Lower energy prices increase the usage of demand response for a given scenario, and
various impacts are improved by higher participation levels. This is the approach the
Hawaiian Electric Company has taken in its demand response program; participants
are given an up-front payment based on their capacity of controllable load (which they
call a capacity payment), but there is no separate payment for each demand response
event that occurs. [18]
Other key findings of the model relate to the characterization of loads:
5. Relative costs to shift or curtail end uses matter.
Much like conventional generation, the optimal dispatch scheduling of responsive loads
depends largely on each end use's costs.
In the model, dishwashers are the most
economical means of delaying consumption; this is why they are dispatched for that
purpose more frequently than other end uses. Similarly, refrigerators and freezers are
the most economical means of advancing consumption forward in time from the status
quo consumption, and are heavily relied upon. This is not solely dependent on the
energy payments required for modification of a given end use, but also on the recovery
penalty. In the model, it requires slightly higher energy payments to shift commercial
HVAC than residential HVAC, but commercial HVAC systems are assumed to operate
112
more efficiently, and the smaller recovery penalty makes commercial HVAC's value
dominate that of residential systems.
6. Diversity of operational constraints of end uses is valuable.
While eight classes of end use were included in the model, some are very similar
in terms of operational constraints, to the point of being nearly redundant (e.g.,
commercial and residential HVAC). Due to these redundancies, only four of the eight
classes of end use were used for load curtailment or shifts in any of the modeled
scenarios (although others were used to provide reserves).
This is because in cases
where certain end uses' constraints were nearly redundant, the end use with relatively
flexible constraints dominates. For example, aside from differences in cost parameters,
residential washer-dryer systems are modeled with identical load shift constraints to
dishwashers, except for a longer load length (reducing the operational flexibility) and
an ability to delay usage by only 3 hours instead of 12.
These changes mean that
laundry devices would only be used if the full capacity of all participating dishwashers
were fully exhausted.
On the other hand, end uses which can fill an operational
niche are highly non-redundant and more valuable.
One example in the model is
refrigerators and freezers which, unlike many of the end uses, offer the ability to bring
consumption forward in time rather than only delaying it.
These findings suggest that in a real demand response program, it would be useful to gain
a fuller understanding of the actual operational constraints and performance parameters
of end uses.
Some of these characteristics are truly technology dependent on the type
of load, as well as the specific model installed by a given customer (e.g., HVAC devices'
cycling durations and recovery penalties). In a highly efficient dispatch scheme for demand
response, a ranking of the participating appliances of each end use type (e.g., a ranking of
the customers with the most and least efficient HVAC units) could be used to set payment
levels and determine dispatch.1 Other characteristics, such as allowable load shift lengths
and curtailment durations, would further depend on demand response program design.
Another significant insight from this model relates to the planning and investment in
CSP with thermal energy storage, or other options for flexibility:
'This does not imply the added administrative costs would justify the gains in operational efficiency for
such a granular ranking of appliances.
113
7. Demand response and storage should be planned in an integrated manner.
In the modeled scenarios, high participation levels for demand response reduce the
usage of CSP's storage; CSP generation is shifted much less often, and is largely
used when solar resources are available.
While the storage of CSP still holds an
inherent value in mitigating intermittency and providing a flexibility option, optimal
investment in storage may be reduced somewhat if a large-scale demand response
program is in place. The same could be said for storage provided by pumped hydro
systems, or even the flexibility provided by a grid interconnect. Each form of flexibility
has different operational tradeoffs; demand response, like CSP, is not equally available
at any given time, but it also can be shifted without a recovery penalty (in the case
of appliances), unlike some forms of storage.
The introduction of demand response-or, for that matter, the development of other forms
of responsive demand such as dynamic electricity pricing-has the potential to help Cyprus
meet its 16% goal for renewable energy generation, lower operating costs, and curb carbon
emissions. However, each of these benefits depends on the design of the program including
aspects out of the scope of this work, such as customer engagement. Aggregating individual
energy users to better shape the demand curve involves many complexities, a few of which
are discussed here, but the potential benefits to the Cypriot energy sector appear to make
the challenge worthwhile.
114
Appendix A
Abbreviations
Abbreviation
CERA
CCGT
CO 2
CSP
DR
DSW
EAC
EU
GJ
HVAC
ICE
kg
kW
MT
MW
MWh
NG
OCGT
O&M
PV
RTP
ST
s.t.
TES
TSO
WILMAR
Meaning
Cyprus Energy Regulatory Authority
Combined Cycle Gas Turbine
Carbon Dioxide
Concentrating Solar Power
Demand Response
Desalinated Sea Water
Electricity Authority of Cyprus
European Union
Gigajoule
Heating, Ventilation, & Air Conditioning
Internal Combustion Engine
Kilogram
Kilowatt
Metric Ton
Megawatt
Megawatt-Hour
Natural Gas
Open Cycle Gas Turbine
Operations & Maintenance
Photovoltaic
Real-Time Pricing
Steam Turbine
"such that"
Thermal Energy Storage
Transmission System Operator
Wind Integration in Liberalized Markets
Table A.1: Abbreviations Used in the Text
115
116
Appendix B
Solar Data Generation Method
This appendix describes the approach used to convert two days worth of solar resource
profiles per month-one typical day, and one day with maximum sunlight for that monthinto data for the full year. The following algorithm is repeated for each day of the year.
1. For PV input data (global irradiance) only, reduce each hour's solar resource value
by 1.1% per each degree of temperature above 32 Celsius (89.6 Fahrenheit). This is
done for both typical and maximum sunlight conditions, as the two profiles include
separate temperature values.
2. Calculate the day's solar profile in maximum sunlight conditions as a weighted average of the current and adjacent months' maximum sunlight data. For example, the
maximum daily sunlight profile on March 1 would be the average of the February and
March maximum sunlight profiles, whereas that of March 16 would simply be equal
to the March maximum sunlight profile.
3. Calculate a probability per month of no rain on a given day, p(no rain), using as a
reference point the number of rainy days per month in Limassol [67].
4. Calculate the probability per month of a day with maximum sunlight, p(full sun), as
p(full sun) = 0.4 x p(no rain)2 . This calculation is needed because available historical weather data does not distinguish between sunny and cloudy days. The chosen
coefficient and squaring of probabilities were tuned to yield an annual capacity factor
similar to that estimated in Cyprus's renewable energy action plan [22].
117
5. Generate a random number between 0 and 1, rand. Depending on the value of rand,
assign the day's solar inputs as follows:
Some rain: rand < p(no rain)
The day's irradiation profile is scaled somewhere between 60% and 100% of the
typical day's data depending on the value of rand. The magnitude of this "rain
penalty" was tuned to help yield an appropriate annual capacity factor.
No rain, but less than maximum sun: p(no rain) < rand < p(full sun)
The day's irradiation profile is scaled somewhere between the typical day's data
and the day's maximum sunlight profile, depending on the value of rand.
Maximum sun: p(full sun) < rand
The daily solar inputs are equal to the day's maximum sunlight profile.
118
Appendix C
Additional Figures
C.1
Figures for 2020 with Natural Gas Generation and Fast
Demand Growth
119
Larnaca CSP Stirling 50.8 MW
Nicosia CSP Stirling -25.5 MW
Solar PV 192.0 MW
Wind -300.0 MW
Pentakomo CSP DSW 4.5 MW
Akrotiri CSP Trough 50.0 MW
Vasilikos CCGT 3
Vasilikos CCGT 2
Vasilikos CCGT 1
Dhekelia ICE 2-3
Dhekelia ICE 2-2
Dhekelia ICE 2-1
Dhekelia ICE 1-3
Dhekelia ICE 1-2
Dhekelia ICE 1-1
Vasilikos Steam 3
Non-fuel variable costs
Fuel cost at maximum load
Extra fuel cost at minimum load
Vasilikos Steam 2
Vasilikos Steam 1
DR Res. Dishwasher
DR Res. Washer-Dryer
DR Res. Washer
DR Desal
DR Comm. HVAC
DR Res. HVAC
DR Res. Refrig-Freezer
DR Comm. Curtailments
Dhekelia Steam 6
Dhekelia Steam 5
Dhekelia Steam 4
Dhekelia Steam 3
Dhekelia Steam 2
Vasilikos CCGT 1 Peak
Vasilikos OCGT 1
Moni OCGT 4
Moni OCGT 3
Moni OCGT 2
Moni OCGT 1
0
100
200
400
300
Variable costs (euros/MWh)
500
600
700
Figure C-1: Variable Costs Per Unit
(2020, Natural Gas, Fast Demand Growth, DR Price=50%, DR Participation=50%)
120
CCGT
ICE I-
I ST
Peaker
CSP+TES
CSP
I
]
PV
Wind
120 0
100
.. .. .
0
80
0
0
40 0
60
0
0
20 0
20
40
15
10
5
-
0
15
10
50
0
0
100
15
0
-
/
--. -. -.
. .
-.
0
.- .-
A
.. .. .
5
15C
0
1000
500
0
12AM, 01/24
12AM, 01/25
12AM, 01/26
12AM, 01/27
12AM, 01/28
12AM, 01/29
12AM, 01/30
12AM, 01/31
Time
Figure C-2: Unit Commitment in Baseline Case, 1/24-1/30
(2020, Natural Gas, Fast Demand Growth, DR Price=oo, DR Participation=0%)
121
ICE
CCGT
ST
CSP
CSP+TES
[~
PV
Wind
DR
1200
1000
800
600
400
200
0
10
5
y-y
5
0
150
100
a)
50
0
a)
18
0
.. ... .. . .. . .. . . .. ... . .
. ... . . .. . . . .. . . . . . . . ...
0
.
.
0 b
VTA
L......
. -..
.. .
-. .. .
I
150
100
50
A
(I
-j -
i
k
0
15
10
50
.. ..
-.
..
..-....
0
150C
1000
500
0
12AM, 01/24
12AM, 01/25
12AM, 01/26
12AM, 01/27
12AM, 01/28
12AM, 01/29
12AM, 01/30
12AM, 01/31
Time
Figure C-3: Unit Commitment with Demand Response, 1/24-1/30
(2020, Natural Gas, Fast Demand Growth, DR Price=50%, DR Participation=50%)
122
CCGT
ICE L
ZST
CSP [
CSP+TES
-
PV
Wind
1500
1000
500
150
100
.
50
A......................
0
200
-
A
-
I
150
100
.
V........
./
~.
I
50
~
~.
0
\.
.
.'I
I
150m
100
50
0A
1500
1000
500
0
12AM, 07/18
12AM, 07/19
12AM, 07/20
12AM, 07/21
12AM, 07/22
12AM, 07/23
12AM, 07/24
12AM, 07/25
Time
Figure C-4: Unit Commitment in Baseline Case, 7/18-7/24
(2020, Natural Gas, Fast Demand Growth, DR Price=oo, DR Participation=0%)
123
ST
ICE
CCGT
E
CSP+TES
CSP
Wind
PV
_
DR
1500
.. . .. .
1000
..
500
0
10
0
..
........ ...... ..........-.
5
V A, A A -
-5
VAA
VA-
-
-
. .
-10
150
100
9
50
-
-
0
Q 200
01
150
-
100
-
-
..
50
0
150
100
LE.
50
...........
0
-
1500
1000
500
0
-
12AM, 07/18
12AM, 07/19
12AM, 07/20
12AM, 07/21
12AM, 07/22
12AM, 07/23
12AM, 07/24
12AM, 07/25
Time
Figure C-5: Unit Commitment with Demand Response, 7/18-7/24
(2020, Natural Gas, Fast Demand Growth, DR Price=50%, DR Participation=50%)
124
.............
...............
I.. .......
....
......
........
..
MW
1/1/2020 .
.
.
.-
.
10
8
3/1/2020-
5/1/2020
6
4
F
2
0
71/1/2020 [
-2
9/1/2020
-
-4
-6
11/1/2020 [
-8
12/31/2020 L
00:00
-10
06:00
12:00
Time of Day
18:00
24:00
Figure C-6: Impact of Curtailment on Demand by Hour
(2020, Natural Gas, Fast Demand Growth, DR Price=50%, DR Participation=50%)
0
-1
-3
-4
0
S-6
-7
Load Reductions
0
1000
7000
5000
6000
3000
4000
2000
Hours, sorted by net demand level before DR
8000
Figure C-7: Impact of Curtailment on Load Duration Curve
(2020, Natural Gas, Fast Demand Growth, DR Price=50%, DR Participation=50%)
125
-
...............
....
............. ...........
-
..
....
.......
-.-. .......
.........
.....
8
6
-
4
1
2
1,
0
1
II.
'III
ill'
!I
Is .1.1ji.1111
0
I-1 11
-2
II.II II -I
11
ilh
-4
-6
-8
Load Increases
Load Reductions
-10
0
1000
7000
5000
6000
4000
2000
3000
Hours, sorted by net demand level before DR
8000
Figure C-8: Impact of Load Shifts on Load Duration Curve
(2020, Natural Gas, Fast Demand Growth, DR Price=50%, DR Participation=50%)
MW
1/1/2020
.
10
8
3/1/2020
6
4
5/1/2020
r
2
0
7/1/2020
-2
9/1/2020
-4
-6
11/1/2020
-8
12/31/2020
00:00
06:00
12:00
Time of Day
-
-
18:00
24:00
-10
Figure C-9: Impact of Load Shifts on Load Demand by Hour
(2020, Natural Gas, Fast Demand Growth, DR Price=50%, DR Participation=50%)
126
MW
12AM
Day 1
N
I
100
I
6AM
Delayed
LonAd
Day 1
12PM
Appliance
Load Shifts
Day 1
-D
-4
C
50
6PM
0
Advanced
Load
Day 1
-50
Thermal Load
12AM
Day 2
12
.A
, Preconditioning
r,,],
,
PM 6PM
Da Y0 Day 0
,
I
12AM
6AM
12PM
6PM
12AM
6AM
12PM
Day1
Day1
Day1
Day1
Day2
Day2
Day2
Hours Impacted by Load Reduction at Time
-100
r
Figure C-10: Cumulative Impact of Load Shifts Per Start Time of Load Reduction
(2020, Natural Gas, Fast Demand Growth, DR Price=50%, DR Participation=50%)
I
200
100 -
0
-100
-
-200
-
-300
-
-
400
-5
Load Increases
Load Reductions
0
5
10
Hours Offset from Start of Load Reduction
15
Figure C-11: Aggregated Cumulative Operating Mode of Load Shifts
(2020, Natural Gas, Fast Demand Growth, DR Price=50%, DR Participation=50%)
127
Residential Dishwashers
Residential Refrigerators & Freezers
20
200
0
100
S
-20
[
0
0
-40
-
0
rz~0
-100
200
-300
-5
Load Increases
Load Reductions
10
5
0
Hours Offset from Start of Load Reduction
15
Load Increases
Load Reductions
10
5
0
Hours Offset from Start of Load Reduction
-5
Commercial HVAC
15
Desalination
0.4
6
4
0.2k
2
0
-2
-0.2
Load Increases
Load Reductions
-4
-6
-0.4
10
5
0
Hours Offset from Start of Load Reduction
15
-5
Load Increases
Load Reductions
10
5
0
Hours Offset from Start of Load Reduction
Figure C-12: Cumulative Operating Mode of Active Load Shift Programs
(2020, Natural Gas, Fast Demand Growth, DR Price=50%, DR Participation=50%)
128
15
1
0.5
Akrotiri CSP Trough
-
Pentakomo CSP DSW
'
1\
1
0.5
0
NI
V.
0
0
0.4
1
0
0.4
Dhekelia ICE 1 1
7 1\
1
0.5
I
1
-
0.5
0
0
0
0.59
1
Dhekelia ICE 2 1
1
1
0.5
0.5
A
0
0.59
0
1
0.59
0
Dhekelia ICE 1 3
0
1
Dhekelia ICE 1 2
0
1
F.
OL
0
Dhekelia ICE 2 2
1
0.59
Dhekelia ICE 2 3
C.)
0
TN'
0.5
0.5
0L
Ce
0.59
C)4
1
0.5
0
1
0.5
0
r
-0
F
0
01
1
Vasilikos CCGT 1
0
ILII\1\
Vasilikos CCGT 2
~1
0.5
-
-
01
0.55
0
Vasilikos CCGT 3
1
0
0.55
1
0
0.46
1
Vasilikos Steam 3
1
[.
0
-0.55
0.5
1
0
1
Vasilikos Steam 1
Vasilikos Steam 2
0.5
1
0.59
0.5
0.5
-- 0.46
S0
1
Load Factor
-0
0.46
1
Baseline
DR Price = 50%,
DR Participation = 50%
Min load
Figure C-13: Impact of Demand Response on Load Factors of Dispatched Units
(2020, Natural Gas, Fast Demand Growth, DR Price=50%, DR Participation=50%)
129
25.0 21.62
20.0
a)
CU
15.0
-
.30% Participation
a)
N
10.0
M50% Participation
.70% Participation
0
-
5.0
1.00
0.12
-------
0.0
30%
0.27
70%
50%
DR Price
Figure C-14: Normalized Average Daily Demand Response Usage
(2020, Natural Gas, Fast Demand Growth, Deterministic Simulations)
1.4 ,
S..
a)
1.2 -
toC
0
1.0
a)
2
a)
0
S..
Z)
0)
a)
0.8
-
0.
. 30% Participation
0.6
S.C
*0
a)
N
0
0.4
M50% Participation
0 70% Participation
-
0.2-
0.0 30%
50%
70%
DR Price
Figure C-15: Normalized Increase in Efficiency Score
(2020, Natural Gas, Fast Demand Growth, Deterministic Simulations)
3.0 U)
a)
>
2.5
2.0 -
0)~a
75
1.5 -
0 30% Participation
P 50% Participation
N1.01.0
0
.70% Participation
0.5 0.0
30%
50%
70%
DR Price
Figure C-16: Normalized CO 2 Emissions Reduction
(2020, Natural Gas, Gradual Demand Growth, Deterministic Simulations)
130
C.2
Figures for 2020 with Fuel Oil Generation and Fast Demand Growth
Larnaca CSP Stirling -50.8 MW
Nicosia CSP Stirling -25.5 MW
Solar PV -192.0 MW
Wind
Non-fuel variable costs
Fuel cost at maximum load
Extra fuel cost at minimum load
-300.0 MW
Pentakomo CSP DSW 4. MW
Akrotiri CSP Trough 50.0 MW
DR Res. Dishwasher
DR Res. Washer-Dryer
DR Res. Washer
DR Desal
Dhekelia ICE 2-3
Dhekelia ICE 2-2
Dhekelia ICE 2-1
Dhekelia ICE 1-3
Dhekelia ICE 1-2
Dhekelia ICE 1-1
DR Comm. HVAC
DR Res. HVAC
Vasilikos Steam 3
Vasilikos Steam 2
Vasilikos Steam 1
DR Res. Refrig-Freezer
Vasilikos CCGT 3
Vasilikos CCGT 2
Vasilikos CCGT 1
DR Comm. Curtailments
Dhekelia Steam 6
Dhekelia Steam 5
Dhekelia Steam 4
Dhekelia Steam 3
Dhekelia Steam 2
Vasilikos CCGT 1 Peak
Vasilikos OCGT 1
Moni OCGT 4
Moni OCGT 3
Moni OCGT 2
Moni OCGT 1
0
100
200
400
300
Variable costs (euros/MWh)
500
600
700
Figure C-17: Variable Costs Per Unit
(2020, Fuel Oil, Fast Demand Growth, DR Price=50%, DR Participation=50%)
131
ST
ICE
IJ
CSP
CSP+TES
CCGT
Wind
PV
1200
1000
800
600
400
200
0
200
0
L
150
100
50
0
-
--
--
.. . .. . .
. -.
.
-
-
..
. . .... . I.. ..
-.-.-.
.. .-.-.-.. ... . . . . .. .. . . .. . .
-
-
- -
..-.-.-.-
0
150
..
-.
- .....-
0
100
[N
50
*
.
/
15 0
I0
10
0
- --
/
- -
--
-
5
0
150
0
1000
500
0
12AM, 01/24
12AM, 01/25
12AM, 01/26
12AM, 01/27
12AM, 01/28
12AM, 01/29
12AM, 01/30
12AM, 01/31
Time
Figure C-18: Unit Commitment in Baseline Case, 1/24-1/30
(2020, Fuel Oil, Fast Demand Growth, DR Price=oo, DR Participation=0%)
132
ICE
I ST
CCGT
Peaker
CSP+TES
E
CSP
PV
Wind
1200
1000
-
800
600
400
200
0
10
5
A
A
A
......
A
-10
200
150
-
100
50
0
P-4
10
150
50
........
100
50
1500
1000
500
0
12AM, 01/24
12AM, 01/25
12AM, 01/26
12AM, 01/27
12AM, 01/28
12AM, 01/29
12AM, 01/30
12AM, 01/31
Time
Figure C-19: Unit Commitment with Demand Response, 1/24-1/30
(2020, Fuel Oil, Fast Demand Growth, DR Price=50%, DR Participation=50%)
133
DR
ICE
ST
CCGT
CSP+TES
CSP E:=
PV
Wind
1500
1000
500
0
150
-
100
50
AL1_A
0
200
-
150
* /
100
50
*
0
1000
I
LALALALLi
1500
10001500
0
/1
12AM, 07/18
12AM, 07/19
12AM, 07/20
12AM, 07/21
12AM, 07/22
12AM, 07/23
12AM, 07/24
12AM, 07/25
Time
Figure C-20: Unit Commitment in Baseline Case, 7/18-7/24
(2020, Fuel Oil, Fast Demand Growth, DR Price=oo, DR Participation=O%)
134
ICE
ST
CCGT
CSP+TES
CSP
PV
Wind
DR
1500
1000
..
.
500
0
4
2
0
-2
-4
150
C
100
....
.........
*A
.. A-
1~
50
0
0
8 200
0
...
....
--- ..... ...
150
.......
........
100
50
0
150
.
. .
. . . . . .
100
50
0
1500
1000
-
500
0
12AM, 07/18
12AM, 07/19
12AM, 07/20
12AM, 07/21
12AM, 07/22
12AM, 07/23
12AM, 07/24
12AM, 07/25
Time
Figure C-21: Unit Commitment with Demand Response, 7/18-7/24
(2020, Fuel Oil, Fast Demand Growth, DR Price=50%, DR Participation= 50%)
135
MW
10
1/1/2020
8
3/1/2020
6
4
_--H
5/1/2020
-
t
-
7/1/2020
2
0
-2
9/1/2020
-4
-- -o m
-- ....
-6
11/1/2020 [
-8
-7=
-10
12/31/2020
06:00
)0:00
12:00
Time of Day
18:00
24:00
Figure C-22: Impact of Curtailment on Demand by Hour
(2020, Fuel Oil, Fast Demand Growth, DR Price=50%, DR Participation=50%)
0
IIII
-1
I-2
-3
Q)
-4
0
4-5
..
z
-6
-7
Load Reductions
-8
0
1000
7000
6000
5000
4000
3000
2000
Hours, sorted by net demand level before DR
8000
Figure C-23: Impact of Curtailment on Load Duration Curve
(2020, Fuel Oil, Fast Demand Growth, DR Price=50%, DR Participation=50%)
136
10
8
-1 -- 1I
I1
6
11
4
2
Q)
0
41
0
-2
i-F''
-4
z
-6
-8
Load Increases
Load Reductions
-10
0
1000
2000
3000
4000
5000
6000
7000
Hours, sorted by net demand level before DR
8000
Figure C-24: Impact of Load Shifts on Load Duration Curve
(2020, Fuel Oil, Fast Demand Growth, DR Price=50%, DR Participation=50%)
MW
1/1/2020
2-
1
1
__om
10
8
3/1/2020
6
4
5/1/2020 [
--
2
--
7/1/2020
-
~~---- --
0
-
-2
9/1/2020
-4
---
11/1/2020
~
-6
F
-8
12/31/2020 .
00:00
.
.
.
06:00
12:00
18:00
--
24:00
-10
Time of Day
Figure C-25: Impact of Load Shifts on Load Demand by Hour
(2020, Fuel Oil, Fast Demand Growth, DR Price=50%, DR Participation=50%)
137
MW
12AM
Day 1
.
.
100
L-
0
6AM
Day 1
Delayed
Load
50
Appliance
0
Load Shifts
12PM
Day 1
0
E10
t
C2
Advanced
6PM
Day 1
Thermal Load
Preconditioning
Cd 12AM
-4--
Day 2 ~I
12 PM
Da'y0
-50
Load
r
I
6PM
Day 0
12AM
Day1
6AM
Day1
12PM
Day1
6PM
Day1
12AM
Day2
6AM
Day2
12PM
Day2
-100
Hours Impacted by Load Reduction at Time r
Figure C-26: Cumulative Impact of Load Shifts Per Start Time of Load Reduction
(2020, Fuel Oil, Fast Demand Growth, DR Price=50%, DR Participation=50%)
600
400
[
200
0
-200
-400
-600
-800
-1000
-1200
Load Increases
Load Reductions
-1400
-5
0
5
10
15
Hours Offset from Start of Load Reduction
Figure C-27: Aggregated Cumulative Operating Mode of Load Shifts
(2020, Fuel Oil, Fast Demand Growth, DR Price=50%, DR Participation=50%)
138
Residential Refrigerators & Freezers
Residential Dishwashers
100
500
03-100
0
-200
500
-300
Load Increases
Load Reductions
Load Increases
Load Reductions
-400
-5
0
5
10
Hours Offset from Start of Load Reduction
15
-5
Commercial HVAC
0
5
10
Hours Offset from Start of Load Reduction
15
Desalination
100
10
50
0
0
-10
-50
-20
-100
-30
-150
Load Increases
Load Reductions
-40
5
0
5
10
Hours Offset from Start of Load Reduction
Load Increases
Load Reductions
-200
15
-5
0
10
Hours Offset from Start of Load Reduction
Figure C-28: Cumulative Operating Mode of Active Load Shift Programs
(2020, Fuel Oil, Fast Demand Growth, DR Price=50%, DR Participation=50%)
139
15
Pentakomo CSP DSW
Akrotiri CSP Trough
1
-
1
50M4v
'
0.5
0.5
0
1
0.4
&
A
0
0
1
0.4
0
Dhekelia ICE 1 2
Dhekelia ICE 1 1
17MW
1
0.5
0
0.59
0
Dhekelia ICE 1 3
1
Dhekelia ICE 2 1
17MW
0.5
0.5
0
0
0
0
0.59
Dhekelia ICE 2 2
-0
1
17MW
1
1
W
0.59
0
1
Dhekelia ICE 2 3
1 7NMW
1
0.5
0.5
0
0.59
0
-
0.59
0
1
0
I
-
1
F
0
Vasilikos CCGT 1
17AMW
1
0.59
Vasilikos CCGT 2
1
22 ()VJ
.1
1
0.5
0
0
0.55
0.55
Vasilikos Steam 1
1
Vasilikos CCGT 3
1
0.5
0.5
0
0
0
0.55
1
1.3fL 1\\
II
1
0.46
Vasilikos Steam 3
0
Vasilikos Steam 2
1
0.5
I
0
0.46
1
.
0.46
0
Load Factor
- - -
1
I
Baseline
DR Price = 50%,
DR Participation = 50%
Min load
Figure C-29: Impact of Demand Response on Load Factors of Dispatched Units
(2020, Fuel Oil, Fast Demand Growth, DR Price=50%, DR Participation=50%)
140
3.0 2.5
2.40
-
1.5 N
E
0
C4.
1.0 -
M 70% Participation
0
0.5
30% Participation
0 50% Participation
1.00
-
0.08
0.18
0.0 1
50%
30%
70%
DR Price
Figure C-30: Normalized Average Daily Demand Response Usage
(2020, Fuel Oil, Fast Demand Growth, Deterministic Simulations)
1.4 1.2 -
0)
0
1.0 -
0)
0)
0
b-
0.8
a6
0
11
u 0.6
-
M30% Participation
Sm50%
0)
N
0
0.4
-
0.2
-
Participation
M 70% Participation
0.0
30%
50%
70%
DR Price
Figure C-31: Normalized Increase in Efficiency Score
(2020, Fuel Oil, Fast Demand Growth, Deterministic Simulations)
141
2.5 2.22
0
2.0
5 1.5
M30% Participation
0.97
1.00
M 50% Participation
1.0 -
M70% Participation
U~
0.5
-
0.29
0,
0.0
30%
70%
50%
DR Price
Figure C-32: Normalized CO 2 Emissions Reduction
(2020, Fuel Oil, Fast Demand Growth, Deterministic Simulations)
142
Bibliography
[1] Giamouridis A. The offshore discovery in the Republic of Cyprus: Monetisation
prospects and challenges. Technical report, Oxford Institute for Energy Studies, 2012.
[2] United States Energy Information Administration. Annual energy outlook 2013 early
release overview, December 2012.
[3] Siemens AG. Steam turbines for CSP plants. Technical report, 2011.
[4] International Energy Agency. Gas medium-term market report 2012: Market trends
and projections to 2017. Technical report, 2012.
[5] International Energy Agency. Golden rules for a golden age of gas: World energy
outlook special report on unconventional gas. Technical report, 2012.
[6] International Energy Agency. Energy prices in national currency per unit, 2013.
[7] Houda Ben Jannet Allal. Combined solar power and desalination plant: Technoeconomic potential in Mediterranean partner countries, 2010. European Commission
MED-CSD Project.
[8] Jason W. Black. Integrating Demand into the U.S. Electric Power System: Technical,
Economic and Regulatory Frameworksfor Responsive Load. PhD thesis, Massachusetts
Institute of Technology, 2005.
[9] S. Borenstein. The long-run efficiency of real-time electricity pricing, 2005.
[10] L.H. Fink C. Concordia and Poullikkas G. Load shedding on an isolated system. IEEE
Transactions on Power Systems, 10(3):1467-1472, 1995.
[11] P. Fokaidis C. Koroneos and N. Moussiopoulos. Cyprus energy system and the use of
renewable energy sources. Energy, 30:1889-1901, 2005.
[12] Shantanu Chakraborty, Tomonobu Senjyu, Ahmed Yousuf Saber, Atsushi Yona, and
Toshihisa Funabashi. Optimal thermal unit commitment integrated with renewable
energy sources using advanced particle swarm optimization. IEEJ Transactions on
Electrical and Electronic Engineering,4(5):609-617, 2009.
[13] European Commission. Cyprus: Internal market fact sheet, 2007.
[14] Federal Energy Regulatory Commission. A national assessment of demand response
potential. Staff Report, 2009.
143
[15] Hawaiian Electric Company. Hawaiian Electric Company demand response: Fast demand response basics.
[16] Hawaiian Electric Company. Fact sheets: Power facts, 2011.
[17] Hawaiian Electric Company. Business Demand Response overview: Energy solutions
for business, 2012.
[18] Hawaiian Electric Company. Demand response overview, 2012.
[19] C. Concordia, L.H. Fink, and G. Poullikkas. Load shedding on an isolated system.
Power Systems, IEEE Transactions on, 10(3):1467-1472, 1995.
[20] Infinia Corporation. Infinia Corp. partner in two NER300 projects awarded by the
European Commission. Press Release, December 2012.
[21] D. Karl Critz. Power system balancing with high renewable penetration: The potential
of demand response. Master's thesis, Massachusetts Institute of Technology, 2011.
[22] Industry Cyprus Ministry of Commerce and Tourism. Cyprus national renewable energy action plan. Technical report, 2009.
[23] Sarah Busche D. Karl Critz and Stephen Connors. Power systems balancing with high
penetration renewables: The potential of demand response in Hawaii. Technical report,
Joint Institute for Strategic Energy Analysis, 2012. Working Paper.
[24] C.S. loakimidis D. Livengood, F.C. Sim-Sim and R. Larson.
isolated energy systems. Island Sustainability,130:197, 2012.
Responsive demand in
[25] Pamela DeAmicis. Seasonal and diurnal variability of wind and hydro energy sources
on the Azores, Portugal and the effectiveness of utilizing energy storage to achieve
maximum penetration. Master's thesis, Massachusetts Institute of Technology, 2011.
[26] E. Denny, A. Tuohy, P. Meibom, A. Keane, D. Flynn, A. Mullane, and M. O'Malley.
The impact of increased interconnection on electricity systems with large penetrations of wind generation: A case study of Ireland and Great Britain. Energy Policy,
38(11):6946 - 6954, 2010.
<ce:title>Energy Efficiency Policies and Strategies with
regular papers.</ce:title>.
[27] R.R. Dholakia and N. Dholakia. From social psychology to political economy: A model
of energy use behavior. Journal of Economic Psychology, 3:231-247, 1983.
[28] P. Meibom R. Barth A. Tuohy E. Ela, M.MIlligan. Advanced unit commitment strategies for the U.S. Eastern Interconnection. 9th Annual International Workshop on
Large-Scale Integration of Wind Power into Power Systems as well as on Transmission
Networks for Offshore Wind Power Plants, October 2010.
[29] Antonis Tsikalakis Emmanouil Psychogyiopoulos and Nikos Hatziargyriou. Short-term
scheduling of desalination loads in hybrid systems. School of Electrical and Computer
Engineering, National Technical University of Athens.
[30] M. Eslamian, S.H. Hosseinian, and B. Vahidi. Bacterial foraging-based solution to the
unit-commitment problem. Power Systems, IEEE Transactions on, 24(3):1478-1488,
2009.
144
[31] Eurostat. Electricity prices for household consumers (EUR per kWh), February 2013.
[32] W.H. Golove and J.H. Eto. Market barriers to energy efficiency: A critical reappraisal
of the rationale for public policies to promote energy efficiency, 1996.
[33] Yakov Hain. Euro-asia interconnector project: Submarine power cable. feasibility study
status executive summary. In Doing Business with Israel Summit. Israel Electric Corporation Ltd., 2012.
[34] A. Hajipapas and K. Hope. Protests follow Cyprus navy fire deaths, July 2011.
[35] Charles Hansen and Alex Papalexopoulos. Operational impact and cost analysis of
wind generation in the island of Crete. IEEE Systems Journal, 2011.
[36] The Cyprus Institute. Techno-economic study of the feasibility of cogeneration of
electricity and desalinated sea water using concentrated solar power. Technical report,
The Cyprus Institute, 2010.
[37] Kitsios K. and Theodoros Z. Energy efficiency policies and measures in Cyprus. Technical report, Cyprus Institute of Energy, October 2012.
[38] Eleni Sardianou Konstadinos Abeliotis, Niki Nikolaou. Attitudes of Cypriot consumers
on the ownership of household appliances:the case of the city of Limassol. International
Journal of Consumer Studies, 35(2):132-137, March 2011.
[39] Nasos Ktorides. Euro-asia Interconnector feasibility study status: Introductory speech.
In Doing Business with Israel Summit. DEH Quantum Energy, 2012.
[40] Woei Ling Leow.
Zoning and Occupancy-Moderation for Residential SpaceConditioning Under Demand-Driven Electricity Pricing. PhD thesis, Massachusetts
Institute of Technology, 2012.
[41] Daniel Livengood and Richard Larson. The energy box: Locally automated optimal
control of residential electricity usage. Service Science, 1(1):1-16, 2009.
[42] Georgios Maroulis. Integration of electricity from renewables to the electricity grid and
to the electricity market - National report: Cyprus. Technical report, Eclareon gmbh
and Institute for Applied Ecology, 2011.
[43] P. Meibom, R. Barth, B. Hasche, H. Brand, C. Weber, and M. O'Malley. Stochastic
optimization model to study the operational impacts of high wind penetrations in
Ireland. Power Systems, IEEE Transactions on, 26(3):1367-1379, 2011.
[44] S. Lefton D. Agan N. Kumar, P. Besuner and D. Hilleman. Power plant cycling costs.
Technical report, National Renewable Energy Laboratory, 2012. Prepared by Intertek
APTECH.
[45] T. Nagata, M. Ohono, J. Kubokawa, H. Sasaki, and H. Fujita. A multi-agent approach
to unit commitment problems. In Power Engineering Society Winter Meeting, 2002.
IEEE, volume 1, pages 64-69 vol.1, 2002.
[46] Electricity Authority of Cyprus. New business development: Renewable energy sources
(RES).
145
[47] Electricity Authority of Cyprus.
November 2012.
Consumption and usage cost of home appliances,
[48] Electricity Authority of Cyprus. The cost of electricity generation in Cyprus. News
Release, February 2012.
[49] Electricity Authority of Cyprus. Electricity Authority of Cyprus 2011 annual report,
2012.
[50] Cyprus Transmission System Operator of Electrical Energy. Pumped storage possibilities in Cyprus. In 1st Workshop of the Stories Project, May 2008. Summarized in
Stories Newsletter 2.
[51] Cyprus Transmission System Operator of Electrical Energy. Long term prediction:
Official prediction of total electricity production (Gwh) and power (MW) for the period
2013-2022, 2013.
[52] EuroStat Press Office. Household electricity prices in the EU27 rose by 6.3prices by
12.6News Release, May 2012.
[53] Mancur Olson. The Rise and Decline of Nations: Economic Growth, Stagflation, and
Social Rigidities. Yale University Press, 1982.
[54] Stelios Orphanides. Infinia to supply 25.5 megawatt solar thermal project in Cyprus,
May 2012.
[55] H. V. Larsen et. al. P. Meibom. Wilmar Joint Market Model documentation. Technical
report, Riso National Laboratory, 2006.
[56] M.K. Kakouris et. al. P.A. Pilayachi, N.G. Kalampalikas.
The energy policy of the
Republic of Cyprus. Energy, 34:547-554, 2009.
[57] George Liberopoulos Panagiotis Andrianesis and Constantinos Varnavas. The impact
of wind generation on isolated system: The case of Cyprus. In Powertech Grenoble
2013: Towards Carbon-FreeSociety Through Smarter Grids, 2013.
[58] C.N. Papanicolas. CSP electricity production and desalination (a report from Cyprus).
In Energy, Water and Climate Change - Building Bridges Between Europe and Middle
East/North Africa. The Cyprus Institute, December 2012.
[59] Maximilian Parness. The environmental and cost impacts of vehicle electrification in
the Azores. Master's thesis, Massachusetts Institute of Technology, 2011.
[60] George Partasides. Implementing the RES directive: The Cyprus action plan, February
2010. Ministry of Commerce, Industry & Tourism, Republic of Cyprus.
[61] George Psyllides. Fed up consumers call for end to EAC monopoly, March 2012.
[62] P. Meibom R. Barth, H. Brand and C. Weber. A stochastic unit-commitment model
for the evaluation of the impacts of integration of large amounts of intermittent wind
power. In 9th International Conference on Probabilitstic Methods Applied to Power
Systems. KTH, June 2006.
146
[63] Statistical Service Republic of Cyprus, Ministry of Finance. Census of Population 2001,
February 2004.
[64] Nick Rodriguez Rick Tidball, Joel Bluestein and Stu Knoke. Cost and performance assumptions for modeling electricity generation technologies. Technical Report
NREL/SR-6A20-48595, National Renewable Energy Laboratory, 2010. Prepared by
ICF International.
[65] H. Sasaki, M. Watanabe, J. Kubokawa, N. Yorino, and R. Yokoyama. A solution
method of unit commitment by artificial neural networks. Power Systems, IEEE Trans-
actions on, 7(3):974-981, 1992.
[66] J. Schleich and R. Betz. Incentives for energy efficiency and innovation in the European
emission trading system. In ECEEE 2005 Summer Study - What Works and Who
Delivers?, 2005. Panel 7: New Economic Instruments.
[67] Cyprus Meteorological Service. Limassol precipitation statistics for the period 1991-
2005.
[68] S.N. Singh, J. stergaard, and J. Yadagiri. Application of advanced particle swarm optimization techniques to wind-thermal coordination. In Intelligent System Applications
to Power Systems, 2009. ISAP '09. 15th InternationalConference on, pages 1-6, 2009.
[69] Walter L. Snyder, H.David Powell, and John C. Rayburn. Dynamic programming
approach to unit commitment. Power Systems, IEEE Transactions on, 2(2):339-348,
1987.
[70] A. Soares, A. Gomes, and C.H. Antunes. Domestic load characterization for demandresponsive energy management systems.
In Sustainable Systems and Technology
(ISSST), 2012 IEEE InternationalSymposium on, pages 1-6, 2012.
[71] K. Spees and L. Lave. Demand response and electricity market efficiency. Electricity
Journal,20(3):69-85, 2007.
[72] Ingo Stadler. Power grid balancing of energy systems with high renewable energy
penetration by demand response. Utilities Policy, 16:90-98, 2008.
[73] G.J. Stigler. The theory of economic regulation. The Bell Journal of Economics and
Management Science, 2(1):3-21, 1971.
[74] Stelios Stylianou. The future development of the Cyprus electricity market on conventional and renewable energy sources. In 2nd Gulf Intelligence Levant Energy Forum.
Electricity Authority of Cyprus, 2012.
[75] Zachariadis T. and Pashourtidou N. An empirical analysis of electricity consumption
in Cyprus. Energy Economics, 29:183-189, 2007.
[76] Chun tian Cheng, Sheng li Liao, Zi-Tian Tang, and Ming yan Zhao. Comparison of
particle swarm optimization and dynamic programming for large scale hydro unit load
dispatch. Energy Conversion and Management, 50(12):3007 - 3014, 2009.
[77] Nat. Treadway. Ecopinion survey report: Consumer preferences and the elusive magic
of dynamic pricing. Technical report, Distributed Energy Financal Group LLC, 2013.
147
[78] A. Tuohy, P. Meibom, E. Denny, and M. O'Malley. Unit commitment for systems with
significant wind penetration. Power Systems, IEEE Transactions on, 24(2):592-601,
2009.
[79] Constantinos Varnavas. Integrating intermittent renewables: Challenges and solutions.
In Energy, Water and Climate Change - Building Bridges Between Europe and Middle
East/North Africa. Electricity Authority of Cyprus, 2012.
[80] Kristof De Vos, Andreas G. Petoussis, Johan Driesen, and Ronnie Belmans. Revision of reserve requirements following wind power integration in island power systems.
Renewable Energy, 50(0):268 - 279, 2013.
[81] W. Winter. Final report: Towards a successful integration of large scale wind power
into European electricity. European Wind Integration Study, 2010.
[82] T. Zachariadis. Forecast of electricity consumption in Cyprus up to the year 2030: The
potential impact of climate change. Energy Policy, 38:744-750, 2010.
[83] Long Zhao and Bo Zeng. Robust unit commitment problem with demand response and
wind energy. In Power and Energy Society General Meeting, 2012 IEEE, pages 1-8,
2012.
[84] F. Zhuang and F.D. Galiana. Unit commitment by simulated annealing. Power Systems, IEEE Transactions on, 5(1):311-318, 1990.
148