Helsinki University of Technology
Department of Engineering Physics and Mathematics
Hannele Holttinen
Hourly wind power variations and their impact on the Nordic power
system operation
Licenciate Thesis
December, 2003
HELSINKI UNIVERSITY OF TECHNOLOGY
DEPARTMENT OF ENGINEERING PHYSICS AND MATHEMATICS
ABSTRACT OF LICENTIATE’S THESIS
Author:
Hannele Holttinen
Department:
Department of Engineering Physics and Mathematics
Major subject:
Advanced Energy Systems
Minor subject:
Energy Economics and Power Plant Engineering
Title:
Hourly wind power variations and their impact on the Nordic power
system operation
Title in Finnish:
Laajamittaisen tuulivoiman tuntivaihtelut
Pohjoismaiden sähköjärjestelmään
Chair:
Tfy-56 Advanced Energy Systems
Supervisor:
Prof. Peter Lund
Instructor:
Doc. Ritva Hirvonen
Abstract:
The variations of wind power production will increase the flexibility needed
in the system, when significant amounts of load is covered with wind power.
When we are studying the incremental effects that varying wind power
production imposes on the power system, it is important to study the system as
a whole: only the net imbalances have to be balanced by the system. Large
geographical spreading of wind power will reduce variability, increase
predictability and decrease the occasions with near zero or peak output.
The goal of this work was to estimate the increase in hourly load following
reserve requirements, based on real wind power production and synchronous
hourly load data in the 4 Nordic countries. As an interim result, proper
statistical properties of large scale wind power production data were looked
for, from the statistical analyses of the data sets available for Nordic countries.
The main conclusions of this study are that the hourly variations of large
scale wind power in the Nordic countries stay 98 % of time inside ± 5 % of
installed capacity. This will be seen as an increase in the hourly load following
requirements of the power system, at the stage when wind power is producing
a significant share of the electrical consumption. At a 10 % penetration level
(wind power production of gross demand) this is estimated as 1.5...4 % of
installed wind capacity, taking into account that load variations are more
predictable than wind power variations.
Number of pages
81p.+app.9 p.
Keywords
Department fills
Approved
Library code
2
ja
niiden
vaikutus
wind power, energy system, hourly
variations
Teknillinen korkeakoulu
Teknillisen fysiikan ja matematiikan osasto
LISENSIAATTITYÖN
TIIVISTELMÄ
Tekijä
Hannele Holttinen
Osasto
Teknillinen fysiikka ja matematiikka
Pääaine
Energiateknologiat
Sivuaine
Energiatalous ja voimalaitostekniikka
Työn nimi
Laajamittaisen tuulivoiman tuntivaihtelut
Pohjoismaiden sähköjärjestelmään
Title in English
Hourly wind power variations and their impact on the Nordic power
system operation
ja
niiden
vaikutus
Professuurin koodi ja Tfy-56 Energiateknologiat
nimi
Työn valvoja
Prof. Peter Lund
Työn ohjaaja
Doc. Ritva Hirvonen
Tiivistelmä
Tuulivoiman vaihtelut lisäävät sähköjärjestelmän tarvitsemaa joustavuutta
siinä vaiheessa, kun huomattava osuus sähkön kulutuksesta tuotetaan
tuulivoimalla. Kun tarkastellaan tuulivoiman aiheuttamia lisävaatimuksia
järjestelmään, on tarkasteltava kokonaisuutta: jokaista tuulivoiman vaihtelua
ei tarvitse säätää, ainoastaan järjestelmän näkemät nettovaihtelut. Tuulivoiman
hajauttaminen eri alueille vähentää tuulivoiman kokonaistuotannon
vaihteluita, parantaa ennustettavuutta sekä vähentää ajanjaksoja jolloin
tuulivoimatuotanto on lähellä nollaa tai huipputehoa.
Tutkimuksen tavoitteena oli arvioida tuulivoiman aiheuttama lisäys
tunnittaisen säätöreservin tarpeeseen perustuen toteutuneisiin tuulivoiman
tuotannon ja sähkönkulutuksen aikasarjoihin Pohjoismaissa. Arvion taustaksi
tehtiin tuulivoimatuotantoaineiston tilastollinen analyysi.
Työn päätuloksena todettiin Pohjoismaiden tuulivoiman tuntivaihteluiden
pysyvän 98 % ajasta välillä ± 5 % asennetusta kapasiteetista. Suurimmat
vaihtelut näkyvät tunnittaisen säätöreservin lisäyksenä sitä enemmän mitä
suurempi osa tuulivoimalla sähköntarpeesta tuotetaan. Kun tuulivoiman osuus
on 10 % vuotuisesta sähkönkulutuksesta, reservitarpeen lisäykseksi arvioitiin
1,5–4 % asennetusta tuulivoimakapasiteetista. Tässä on huomioitu
kuormavaihteluiden tuulivoimaa parempi ennustettavuus.
Sivumäärä
81 s.+liitt.9 s.
Avainsanat
Täytetään osastolla
Hyväksytty
Kirjasto
3
tuulivoima, sähköjärjestelmä,
tuntivaihtelut
Table of Contents
Foreword.....................................................................................................................................6
Nomenclature..............................................................................................................................7
1
2
3
4
Introduction ........................................................................................................................8
1.1
Wind power production ..............................................................................................9
1.2
The area of study: the Nordic power system ..............................................................9
1.3
Power system operation............................................................................................10
The impacts of wind power on the power system ............................................................14
2.1
Previous work ...........................................................................................................15
2.2
The aim of this work.................................................................................................17
Data used in this study......................................................................................................18
3.1
Data handling principles...........................................................................................19
3.2
Data set for Finland ..................................................................................................21
3.3
Data set for Denmark................................................................................................23
3.4
Data set for Sweden..................................................................................................24
3.5
Data set for Norway..................................................................................................24
3.6
Long term yearly production data ............................................................................25
Large scale wind power production..................................................................................26
4.1
Basic statistics of the wind power production data used ..........................................28
4.2
Frequency distributions of wind power production..................................................31
4.3
Seasonal variation of wind power production ..........................................................33
4.4
Diurnal variation of wind power production ............................................................35
4.5
Persistence of wind power production......................................................................37
4.5.1
Duration of calms .............................................................................................37
4.5.2
Peaks of wind power production ......................................................................38
4.6
Correlation of wind power production .....................................................................39
4.7
Short term variations of wind power production......................................................42
4.7.1
The in-hour variations ......................................................................................42
4.7.2
The hourly variations........................................................................................43
4.7.3
Variations for longer time scales ......................................................................48
4.8
5
6
7
8
Predictability of wind power production ..................................................................49
Representative data for large scale wind power production.............................................51
5.1
Representativeness of the study years ......................................................................51
5.2
Representativeness of the geographical spreading of data .......................................52
Wind power production and load .....................................................................................55
6.1
Basic statistics of the hourly load time series...........................................................56
6.2
Correlation of load and wind power .........................................................................58
6.3
Temperature dependence..........................................................................................59
6.4
Instant penetration level of wind power ...................................................................61
6.5
Wind power during peak load ..................................................................................63
6.6
Hourly variations of load..........................................................................................65
Increase in net load variations by wind power .................................................................67
7.1
Wind power increasing the largest hourly variation in the system ..........................69
7.2
Wind power increasing the hourly variations in the system.....................................70
7.3
Wind power increasing the unexpected hourly variations of load ...........................72
Summary and conclusions ................................................................................................75
References ................................................................................................................................78
List of Appendices
1. Wind power production curves, 4 countries and total Nordic production, years 2000-01.
2. Hourly variations of wind power production 4 countries, years 2000-01.
3. Hourly load data, 4 countries and total Nordic load, year 2000.
4. Hourly variations of load, 4 countries and total Nordic load variations, years 2000-01.
5
Foreword
This licenciate thesis has been carried out at Technical Research Centre of Finland VTT, wind
power team1. The work was partly financed by Fortum Säätiö (Fortum Foundation), Nordic
Energy Research and EU, through research project WILMAR (Wind power Integration in
Liberalised electricity markets), with Finnish Energy Industries Federation Finergy cofunding.
First of all I want to thank the wind power producers that have given hourly production data
from their wind parks, as well as power companies that have given wind speed measurement
series, without which this study would not have been possible.
My supervisor professor Peter Lund and my instructor docent Ritva Hirvonen2 have given me
valuable comments for the work, for which I am grateful. This work is a fruit of Nordic cooperation – visiting the research institutes and power companies in Denmark, Norway and
Sweden has given me a better opportunity to obtain data for this study, as well as interesting
discussions on the impacts of wind power on the power system. The wind power team as well
as the energy systems group at VTT has been a good working environment, thank you
colleagues! Special acknowledgement goes to Göran Koreneff, who has made load forecast
estimates for Finland used in this work.
Last but not least, my family has given me the hugs and kisses needed to keep me going.
Special thanks for the patience of my daughters Sara and Meri, not getting angry when the
mother had eaten the last chocolate biscuits. And to my dear husband Esa, for his love and
impatience.
Espoo, December 2003
Hannele Holttinen
1
VTT Processes, Energy production research area, Distributed energy group
docent at Power Systems Laboratory of Helsinki University of Technology, currently working as Head of Unit,
Natural Gas and Electricity Transmission, Energy Market Authority EMA
2
6
Nomenclature
Alphabets
d
distance between two sites
D
decay constant
i
index, hour
I
increase in variations
L
load, electric consumption
n
number of data points in time series
NL
net load (load – wind power production)
p
wind power production relative to installed capacity
P
wind power production
P TOT
nominal power of wind power (installed capacity)
rxy
cross correlation
x
data value in time series 1
y
data value in time series 2
Greek letters
σ
standard deviation
µ
mean value of time series
∆
hourly variation
Abbreviations
DC
direct current
CHP
combined heat and power production
7
1 Introduction
Integration of wind power in large power systems is mainly subject to theoretical studies, as
wind power penetration levels are still modest. Even though the penetration in areas like West
Denmark is already high (about 20 % of yearly electricity consumption), wind power presents
only 1–2 % of Nordel or Central Europe (UCTE) systems.
Wind power production is characterised by variations on all time scales: seconds, minutes,
hours, days, months and years. Even the short term variations are to some extent
unpredictable. These are the main reasons why large scale wind power production poses a
challenge to the rest of the energy system. To what extent this will be a problem, depends on
how large a share is produced by wind power, as well as on the power system in question: the
inherent load variations and flexibility of the production capacity mix.
For the power system, the relevant wind power production to study is that of larger areas. This
means large geographical spreading of installed wind power, which will reduce the variability
and increase the predictability of wind power production. Not taking this into account can
result in an exaggeration of the impacts of wind power. On the other hand, assuming that the
smoothing effect of large scale wind power will take care of all problems, will result in an
underestimation of the impacts of wind power.
This study is one step in the way of quantifying the impacts of large scale wind power on the
operation of power system, based on existing production and load data on an hourly level.
Detailed statistical analyses of hourly wind power production are presented. The aim is to see
how large scale, regional wind power production looks like compared with that of a single
wind farm, and going further, how wind power production from the whole Nordic area looks
like compared with the production from one country only.
After looking at the wind power production on its own, the analysis is broadened to contain
the electricity system. Wind power production is studied together with the electrical demand,
i.e. the load. The aim of the study is to look at the effects of large scale wind power
production on power system operation. The main focus is on the hourly variations of wind
power and their effects on the needs for flexibility for the electricity system.
We start by introducing some basic properties of wind power production, the area of the
study, as well as a short description of power system operation. Chapter 2 describes the
impacts of wind power to power systems, and sets this study into perspective. Chapter 3
presents the data used in this study. Chapter 4 illustrates the patterns and smoothing effect of
large scale wind power production, with a short description of the predictability. Chapter 5
gives an overview of representativeness of large scale wind power data, setting the data used
in this study in perspective. Chapters 6 to 7 deal with the variations of wind power production
seen from the electric system point of view, and chapter 8 concludes the findings of this
study.
8
1.1 Wind power production
Wind power is usually referred to as its nominal installed capacity, the peak power. The
nominal power is usually reached only for less than 5 % of time. For the power system, wind
power can also be seen as production resource with a (low) average power and a large
variation.
Wind power production is highly dependent on the wind resource of the site. Therefore both
the average production, distribution of production, as well as seasonal and diurnal variation
can look very different in different sites and areas of the world. For most sites on land, the
average power as % of nominal capacity is between 20–40 %. This can be expressed as full
load hours of 1800–3500 h/a (full load hours is the annual production divided by nominal
capacity). Offshore wind power production, or some extremely good sites on land, can reach
up to 4000–5000 full load hours (average production 45–60 % of nominal capacity).
Full load hours is a theoretical figure only, for comparison of different power plants, so it
does not tell how many hours the power plant is actually operating. For wind power, which
operates most of the time at less than half of the nominal capacity, the turbines will typically
produce power during 6000–8000 hours a year (70–90 % of time). Also, the full load hours of
wind power can be influenced by the design of wind turbines (generator size compared to
wind turbine rotor swept area). The values given here are the result of standard economic
optimisation of investment costs versus production.
For comparison: for combined heat and power production (CHP) full load hours range
typically between 4000–5000 h/a, for nuclear power 7000–8000 h/a and for coal fired power
plants 5000–6000 h/a.
The available wind resource will vary from year to year. Wind power production during one
year ranges typically between ± 15 % of the average long term yearly production (Ensslin et
al, 2000; Giebel, 2001).
1.2 The area of study: the Nordic power system
The joint, liberalised Nordic electricity market covers Norway, Sweden, Finland and
Denmark.
Eastern part of Denmark is part of the Nordel system, and Western part of Denmark is part of
the Central Europe UCTE system. They are not connected by a transmission line, but are both
connected to Sweden and Germany, and West Denmark also to Norway by a DC link. The
production mix is shown in Fig.1. A large share of hydro power is characteristic for the
Nordic area: Norway covers almost 100 %, Sweden almost 50 % and Finland almost 20 % of
electricity consumption by hydro power.
The installed wind power capacity in the beginning of year 2003 was 2200 MW in West
Denmark, 573 MW in East Denmark, 345 MW in Sweden, 97 MW in Norway and 41 MW in
9
Finland (Eltra, 2003; Elkraft 2003; Carlstedt, 2003; Laakso, 2003; Windpower Monthly,
2003). In Western part of Denmark, system integration of wind power is already reality,
whereas in other countries it is a subject for discussion. In Denmark, the scheduling of
production units takes into account wind power production, and prediction methods together
with the hourly trade in the spot and regulation markets are used in order to accommodate the
substantial share of wind power in the system (Eriksen et al, 2002).
18 %
82 %
8 TWh
20 %
0,7 %
23 %
19 %
55 %
51 %
30 %
6%
99%
122 TWh
71 TWh
44 %
1,6 %
50 %
395 TWh
12 %
88 %
Hydro power
Wind power and geothermal
Nuclear power
Thermal power
158 TWh
36 TWh
Figure 1. Electricity production in the Nordic countries in 2001. (Source: Nordel/Finergy)
1.3 Power system operation
Electric power systems include power plants, consumers of electric energy and transmission
and distribution networks connecting the production and consumption sites. This constantly
fluctuating interconnected system should maintain balance between production and
consumption of electric energy. Faults and disturbances should be cleared with smallest
disadvantage in the delivery of electricity.
Power systems comprise a wide variety of generating plant types, which have a range of
capital and operating costs. The operation of power system involves providing a total amount
of electricity, at each instant, corresponding to a varying load from the electricity
consumption. To make this cost effectively, the power plants running at low operational costs
will be kept running almost all the time (base load demand), and the power plants with higher
costs will be run only when the load is high. When ignoring second order costs (like start-up,
shutdown, reserves), plants can be stacked in merit order, where production with low
marginal costs run first. Wind power plants (as well as other variable sources like solar and
tidal) have very low marginal costs, usually assumed as 0, so they come to the top of the merit
order, that is, their power is used whenever available (Grubb, 1991). The electricity markets
10
operate in a similar way, at least theoretically. The price the producers bid to the market is
slightly higher than their marginal cost, because it is cost effective for the producers to
operate as long as they get a price higher than their marginal costs. When the market is
cleared, the power plants operating at lowest bids come first.
The failure to keep the electricity system running has high and costly consequences, thus the
reliability of the system has to be kept at a very high level. Security of supply needs to be
maintained for both short and long term. This means maintaining both flexibility and reserves
necessary to keep the system operating under a range of conditions, also in peak load
situations. These conditions include credible plant outages, as well as predictable and
uncertain variations in demand and in primary generation resources including wind.
The operational reserves are divided into different categories according to the time scale they
are operating. Roughly, a division to regulation (seconds-minutes; primary reserve) and load
following (from minutes to hours; secondary reserve) can be made. Actually load following is
done partly beforehand as scheduling and dispatch of power plants according to load forecast,
and partly by secondary reserve to balance the load forecast errors. Power system operation
involves also the start-ups and shut downs of slower power plants, called unit commitment, in
the time scale of 3…12 hours.
The operation of the power system has to be guaranteed also in the free electricity markets.
Usually Independent System Operator (ISO) as a system responsible grid company takes care
of the whole system using active and reactive power reserves to maintain system reliability,
voltage and frequency. The scheduling and dispatch of the power plants (unit commitment
and load following according to load forecasts) can be dealed with in the electricity markets,
for example the Nordpool ELSPOT market in the Nordic countries. Contracts between some
producers (and consumers) and system operators are made to allocate the primary and
secondary reserves. The secondary reserves can also be dealed partly in the markets, at a
special regulation market. In the Nordic countries, the system operators maintain a joint
regulation market since 2003 (Nordpool, 2003).
The power system, which is operated synchronously, has same frequency. Frequency shows
balance or imbalance between production and consumption in the power system. With
nominal frequency (in Europe 50 Hz) the production and consumption (including losses in
transmission and distribution) are in balance. When frequency is below 50 Hz the
consumption of electric energy is higher than the production. If frequency is above 50 Hz the
consumption of electric energy is lower than the production. Frequency deviates from the
nominal value the less the better the balance between production and consumption can be
maintained. For example in Nordic Power System, the frequency is allowed to range between
49,9 Hz and 50,1 Hz (Fingrid, 2003). Frequency of the system is maintained within allowed
limits by using primary reserve in power plants. It is activated automatically by frequency
fluctuations.
Fig. 2 shows an example of the actual load in the system during 3 hours compared to hourly
forecasted load denoting forecast errors and short-term load deviations in the system. An
example of how the operational reserves operate, is illustrated in Fig.3. It shows the activation
of reserves and frequency of the system as a function of time when a large power plant is
disconnected from the power system. Activation of reserves divides the reserves into primary
11
reserve, secondary reserve (also called fast reserve) and long-term reserve (also called slow
reserve or tertiary reserve). Secondary reserve is active or reactive power activated in 10 to 15
minutes after the occurrence of frequency deviation from nominal frequency. It replaces the
primary reserve and it will be in operation until long-term reserves substitute it as seen from
Fig.3. The secondary reserve consists mostly of rapidly starting gas turbine power plants,
hydro (pump) storage plants and load shedding. Usually every country in interconnected
power system has a secondary reserve, corresponding to the amount of disconnected power
during the dimensioning fault (usually loss of largest power unit) within the country in
question. To provide sufficient secondary power reserve system operators may take load
forecast errors into account. In this case the total amount of the secondary reserve may reach a
value corresponding to about 1.5 times the largest power unit (Holttinen & Hirvonen, 2004).
Power
Forecast
Short-term load
deviation
Load
(actual)
Forecast error
Time / hours
Figure 2. Example of actual load in the system during 3 hours compared to forecasted load.
Time
Frequency
50 Hz
Frequency dependent
load decrease
0
Load
Secondary
reserve
Power
Kinetic energy
Frequency
Primary reserve
Long term reserve
Time
Minutes
Seconds
Hours
Figure 3. Activation of power reserves and frequency of power system as a function of time when a
large power plant is disconnected from the power system (Source: Hirvonen, 2000).
12
The system works for the consumers, and they also pay the system cost in their tariffs, like
they pay for the production, distribution and taxes. It is not usually necessary to allocate these
system costs to a certain producer or consumer. However, when we are talking about larger
changes in the system, introduced by large scale wind power for example, it is appropriate to
consider also the changes needed for the system, as well as the possible extra costs related to
these changes.
13
2 The impacts of wind power on the power system
Wind speed varies in all time scales, and this has different effects on the power system. Wind
gusts cause second to minute variations, which time scale is relevant for frequency control
(primary reserve). Changing weather patterns can be seen in the hourly time series of wind
power production. This is the time scale from secondary reserve, load following to unit
commitment. Diurnal cycle is present in this time scale, too, whereas the seasonal cycle and
annual variations are relevant for the long term adequacy studies. For system planning,
extreme variations of large-scale wind power production are of importance, together with the
probability of the variations. The system impacts of wind energy are presented schematically
in Fig. 4. These impacts are divided into two: short term, balancing the system in the
operational time scale (minutes to hours), and long term, providing enough power and energy
in peak load situations.
For the operation of power systems, the variations from day to day, hour to hour and minute
to minute are of interest. The stops of the power plants do not impose as large variations as
the wind variability for large scale wind power, where a single turbine is less than 1 % of
capacity (for example, a 2 MW turbine in a country with 1000 MW wind power).
Voltage management: Reactive reserve. WF can provide.
Local or system area. Time scale up to some minutes
Cycling losses: Unoptimal use of thermal/hydro capacity.
System area. Time scale 1…24 hours
SHORT
TERM
EFFECTS
Transmission/distribution losses (or benefits).
System/local area. Time scale 1…24 hours
Reserves: Load following and regulation (WF can provide partly)
System area. Time scale some minutes to one hour
Discarded energy: wind power exceeds the amount system can absorb..
System area. Time scale some hours
LONG
TERM
EFFECTS
System reliability: Adequacy of power (capacity credit of WP)
System area. Time scale one to some years
Figure 4. System impacts of wind power (WP) and wind farms (WF), causing integration costs. Part of
the impacts can be beneficial for the system, and wind power can have a value, not only costs.
When studying power system impacts of wind energy, we are referring to a larger area than
just one wind farm. The relevant system area to look at varies according to the impact studied.
For voltage management, only areas near wind power plants should be considered. There
should be enough reactive reserve in the system during disturbances, but it should mainly be
managed locally. For intra-hour variations, frequency control and load following, the
14
synchronously operated system forms a relevant area. DC links connecting synchronously
operated areas can also be automised to be used for primary power control; their power
reserve capacity is usually, however, only allocated as emergency power supply. For dayahead hourly production, the electricity market is a relevant area: for example Nordic power
market includes countries situated in different synchronous systems. When looking at a large
interconnected area, it has to be taken into account that benefits exist when there are no
bottlenecks of transmission.
When we are studying the incremental effects that varying wind power production imposes on
the power system, it is important to study the system as a whole: power system is there for all
production units and loads, and only the net imbalances have to be balanced by the system.
The overall system reliability should be held on a same level before and after wind power, and
study the requirements for the power system from this perspective.
There are means to reduce the variations of wind power production. Staggered starts and stops
from full power as well as reduced (positive) ramp rates can reduce the most extreme
fluctuations, in magnitude and frequency, over short time scales (Kristoffersson et al, 2002).
This is at the expense of production losses, so any frequent use of these options should be
weighed against other measures (in other production units) in cost effectiveness.
2.1 Previous work
Studies of large scale wind power production, its variability and effects on energy system
have been carried out to some extent in the 1990’s and increasingly in the first years of the
new millennium. The first comprehensive article about the system impacts of wind power was
by Grubb, 1991, considering the UK power system.
The extent of wind power variability has been the subject of several studies. European
meteorological station wind data for one year has been used in two studies, however, not
covering all of the Nordic countries (Giebel, 2001; Landberg, 1997). In the Netherlands, a
conscientious work on analysing wind speed data was done, including variability and
persistence (van Wijk, 1990). In Ireland the variations of dispersed wind power production as
well as diurnal dependence has been studied with 5 year wind speed data across the island
(Hurley & Watson, 2002). For the Nordic countries, a study based on Reanalysis (weather
prediction) long term 12-hourly wind speed data was made (Giebel, 2000) looking at the
longer (12- and 24-hour) variations and correlation of production. For Finland, yearly and
monthly wind power variations were studied in Holttinen et al, 1996 and 3-hourly variations
based on data for 5 geographically dispersed weather stations by Tammelin & Nurmi, 2001.
All the studies above have been based on wind speed data from several geographically
dispersed measurement masts, converting wind speeds first to higher altitude hub height wind
speeds and then to the production of a single wind turbine using a power curve. There are
possible caveats first of all in upscaling the wind to higher altitudes, as the wind profile is
dependent on atmospheric conditions (van Wijk, 1990) and secondly in using a single point
measurement to represent a wind farm stretching from one to several kilometres in
dimensions. Studies based on wind power production, however, are more scarce, due to the
15
fact that large scale wind power production has only started to emerge in the past few years.
There are, however, some studies using wind power production data. In the Netherlands,
variability and persistence of dispersed wind farm data of 250-500 kW turbines (van Zuylen
et al, 1994) confirm most of the wind speed data analyses of van Wijk, 1990, however,
indication of somewhat less variability when using wind turbine data can be seen. In
Germany, statistical analyses from production and measured wind speed data have been made
in conjunction with a comprehensive 10 year follow up project of the 250 MW programme
(ISET, 2002; Ensslin et al, 2000). Annual, seasonal, diurnal and hourly variations are one
result of this activity. Faster measurements of this data were further analysed to look at the
trends of power fluctuations as well as fast regulation needs of wind power (Ernst, 1999). A
study of the smoothing effect and its saturation was made for Northern part of Germany
(Focken et al, 2001). Fast ramp rates (1 second to 1 hour) for a large wind farm have been
recorded in USA (Parsons et al, 2001; Poore & Randall, 2001).
Taking the analyses of wind power production further, there are studies combining the
variations of wind power and electrical load. In Norway, wind speed data was analysed in
conjunction with peak load periods, concluding also a decrease in the variability of
geographically dispersed wind power production (Alm & Tallhaug, 1993). First analyses of
the Nordic data set analysed in this study were presented in (Holttinen, 2002). Statistical
analysis of the variability of wind power together with load variations has been presented in
(Persaud et al, 2000) and (Milligan, 2003).
First experiences from West Denmark and Northern coast of Germany have shown that when
significant amounts of electrical demand are covered with wind power, increased flexibility is
needed in the system. This is first seen as increased transmission with neighbouring countries
or areas (Eriksen et al, 2002; Lund & Münster, 2003; Holttinen & Pedersen, 2003). Studies
for island systems like Crete and Ireland have shown that some of the wind power produced
will have to be discarded if the penetration is more than 10 % of yearly gross demand
(CER/OFGEM, 2003; Giebel, 2001).
As a conclusion of several studies in USA (Milligan et al, 2002; Hirst, 2002), it has become
clear that to estimate the impacts of wind power on the power system, the wind induced
imbalances have to be treated together with aggregated system imbalances. Estimating the
increased reserve requirements have resulted in very small impact on regulation time scale
(Ernst, 1999; Parsons et al, 2003) and increasing impact on load following time scale with
increasing penetration (Milborrow, 2001; Milligan, 2003). The studies mostly suffer from
lack of detailed, representative data for both the large scale wind power production and the
load from the same area, so they give conservative estimates.
Power system studies are often carried out as the system was operating before liberalisation of
electricity markets. As wind power is new to actors in the electricity market, market rules that
treat wind fairly, neither subsidising nor penalising its operation have not yet been developed.
Some methods for looking at how the imbalance due to variability and unpredictability of
wind come into the markets have been suggested, using either an hour-ahead market or intrahour and hourly imbalance payments (Hirst, 2002).
The influence of wind power to the adequacy of power production in the system is not the
goal of this work. However, as some analyses touch upon this subject, a short overview on
16
this work is given here. It has been shown in several studies that when the capacity of a
variable source is small (low system penetration) the capacity value equals that of a
completely reliable plant generating the same average power at times when the system could
be at risk (Milligan, 2000; Giebel, 2001; Peltola & Petäjä, 1990). As the penetration increases,
variable sources become progressively less valuable for saving thermal capacity. According to
studies made, this will be a level of about 10–15 % of installed capacity. Dispersion of wind
power and a positive correlation between wind power and demand increase the value of wind
for the system. At large penetrations, the capacity credit tends towards a constant value, that
is, there is no increase in the capacity credit when increasing the wind power capacity. This
will be determined by the loss-of-load-probability without wind energy and the probability of
zero wind power. (Giebel, 2001). As wind energy correlates only weakly with hydro power
production, wind energy added to the system can have considerably higher energy delivery
value than adding more hydro (Söder, 1999).
2.2 The aim of this work
This work is not covering all impacts of wind power on the power system described in Fig. 4.
It is based on a data set of realised hourly wind power production values from two example
years. The study is mainly concentrating on the extent of wind power hourly variations, and
their impacts on the secondary reserve, or hourly load following reserve of the power system
(Short tem effects --> Reserves in Fig.4). Some of the analyses of hourly data give also
insight to the adequacy of power systems (Long term effects --> System reliability in Fig.4).
A large part of the work includes analysing the large scale hourly wind power production as
such. It is common sense as well as proven by earlier studies, that geographical spreading of
wind power will reduce variability. However, the quantification of this phenomenon is not
straightforward. This is a relevant research topic in itself, needed in order to find out what
kind of input data for wind power should be used when studies of wind power in power
systems are made.
After analysing the example years of wind power data available, the representativeness of the
data is judged. This is done first as to how well the two example years are representing the
average/low/high wind years. Then the wind power production data is judged in how well it
represents a large scale wind power production, that is, can it be upscaled to present a
significant portion of the electricity system without upscaling large variations present in data
of only few wind farms.
Wind power is studied together with data of electricity consumption from the same area and
example years, to see how much the variations of wind power production affect the varying
load the power system sees. That is, how much addition of wind power will increase the
variations that the power system has to deal with in an hourly level. First the variations are
taken as measure of how much extra reserve is required due to wind power. Secondly, an
effort to take into account the different properties of predictability for load and wind
variations is presented.
17
3 Data used in this study
The data used in this study is the measured output of wind power plants and wind parks
(Fig.5). Realised hourly wind power production time series from 4 Nordic countries were
collected. The total electricity consumption from the countries, also as hourly time series, was
obtained to see the effect of wind variations compared to load variations. Some available
hourly temperature time series were used to study the temperature dependence. Data was
collected for years 1999 to 2001. As data from Denmark was only available from 2000
onwards, and data for Norway and Sweden was limited for 1999, most of the analyses were
made for two years of data, 2000-2001.
The advantage of using realised wind power production data is to get the real wind farm
output. When converting wind speed data to power production, there will always be some
error, especially if the single point measurement is to represent a larger wind farm area. There
will also be the effect of technical availability in the data, some of the turbines being serviced
or faulty.
Figure 5. Data for hourly wind power production was available from 21 sites in Finland, 5 sites in
Sweden, 6-12 sites in Norway (the lighter coloured sites only for part of the time) and the aggregated
total production of hundreds of sites in Denmark West and East.
18
As wind power production data is limited in Finland, Norway and Sweden, also hourly wind
speed measurement data was used to complement the production data. An effort was made to
make single measurement point data represent wind farm production.
Nordic data set was formed from the data sets of the 4 countries. The production at each hour
was a simple average of the % of capacity production of the 4 countries. In terms of capacity
this would mean setting for example 3000 MW in each country, a total of 12 000 MW. This is
somewhat theoretical, as Denmark is now dominating the installed wind power, and probably
will be for quite some time. The wind energy potential, however, is probably as large in all
the 4 countries, when taking the offshore wind power potential in Sweden and Finland into
account.
The rest of this chapter describes in detail the wind power data used in this study. It starts
with outlining the data handling principles. Also the aggregation and upscaling done to
represent large scale wind power production in each country is described.
3.1 Data handling principles
The extensive work for gathering data included a check up to make sure that the time shifts
from winter and summertime were taken the same way in all time series collected, to keep all
hourly values synchronous. Also the time zone difference for Finland was taken into account
when outlining the Nordic data.
For wind power production time series in Finland, Sweden and Norway, the available data
presented far less than 100 MW of capacity. This means that these time series had to be
upscaled more than 10-fold, to make a large scale wind power production time series for the
countries. Upscaling the hourly values means upscaling also the hourly variations. Real large
scale wind power production would mean that the output would be smoothened out hundreds
or thousands of turbines situated in tens or hundreds of sites. An example of the problem is
illustrated in Fig. 6, from real data in Denmark. Upscaling data from a single site would give
us a different kind of hourly time series – more pronounced peaks and variations – than the
real, 500 MW data shows. This is why several, geographically dispersed sites were looked for
to make the aggregate time series for the countries. Also, data from single wind speed
measurement points were smoothed out before using as data for a larger wind farm.
The time series of only few turbines were checked for longer downtimes of turbines. This was
done for several sites in Finland, and the one production series for Norway. Upscaling one
wind farm data of 2...8 MW to 50...300 MW means that a large amount of turbines would
suddenly be unavailable simultaneously for a long period. The technical availability of wind
turbines is usually quite high, more than 95 % is reported from Sweden and Germany
(Carlstedt, 2003; ISET, 2002). For hundreds of single turbines, on the average only less than 5
% of the turbines will be unavailable at the same time.
There were 2 wind speed series for Finland and one for Sweden. Most of the data for Norway
was as wind speed time series. The wind speed was converted to wind power production. First
the wind speed was smoothed out by taking a 2-hour-sliding-average for each hour. This
19
smoothened wind speed was converted to power production using an aggregated, multiturbine power curve, Fig.7 (Nørgaard, 2003).
For single turbine data, the same kind of smoothing was done, by first converting back to
wind speed, and then applying the same method as for the wind speed time series (only one
site in Finland).
450000
East Denmark (500 MW)
15 MW wind farm upscaled to 500 MW
400000
350000
300000
250000
200000
150000
100000
50000
0
1
25 49 73 97 121 145 169 193 217 241 265 289 313 337 361 385 409 433 457 481 505 529 553 577 601
Figure 6. An example of wrong upscaling: a single site would see more variations,
peaks and calms
than dispersed, large scale wind power production (here 500 MW, 200 x 100 km2).
Wind turbine power output (kW)
Multiturbine power curve
Original power curve
2200
2000
1800
1600
1400
1200
1000
800
600
400
200
0
0
3
6
9
12
15
18
21
24
27
30
Wind speed (m/s)
Figure 7. To convert the wind speed time series to wind farm power production, a multiturbine power
curve was used, smoothing out the production near the cut-in (3 m/s) and cut out (25 m/s) wind speeds
compared with a single turbine power curve.
The focus in this study is on the variations of wind power production. Basically the error in
the data sets comes from not having tens of aggregated wind farm time series available to
20
represent a combined production of a country. The data handling procedure trying to smooth
out some of the variations in a point wind speed measurement data is artificial and will
introduce some error to the data sets. The smoothing of the wind speeds, by sliding 2 hour
averages, make the most of the reduction of variability. The use of a multiturbine power curve
will mostly affect the time series near the cut in wind speeds, above 22 m/s (Fig. 7). All in all,
as the yearly energy production will remain the same in the time series, this procedure will
mainly affect the variability of the production, and the error is considered small. For Finland
data set, this procedure was done for 15 % of the data (3 time series representing 150 MW of
a total of 1000 MW). For Norway nearly all data was handled this way, so there is probably
error involved, however, as described before, compared to the original error of not having
enough time series data, this procedure will reduce the error, not increase it.
To compare the data sets of different installed capacity, they were represented as relative
production, as % of installed capacity. It could be useful to represent the data relative to
average power instead of maximum power, installed capacity. However, as the average power
is changing from year to year, the nominal power is here chosen as a relative measure:
pi =
Pi
PTOT
,
(1)
where pi is the relative production for hour i as % of capacity, Pi is the production MWh/h for
hour i and PTOT is the installed capacity.
3.2 Data set for Finland
Even though the amount of wind power in Finland, 41 MW, is still modest, the capacity
installed is well spread along the long coastline and Lapland fells. As a courtesy of 10 wind
power producers, and 2 power companies with wind speed measurements in high masts, wind
power production data was available from a total of 55 turbines on 21 sites and wind speed
data was available from 2 sites (in the internet, only Lumituuli, 2003). The data is presented in
Table 1 and Fig. 8.
The maximum distance between the sites is 1000 km North-South and 400 km West-East. For
year 1999, when some of the turbines were erected during the year, first part of the year was
taken from wind speed measurements of the Finnish Meteorological Institute, from the
synoptic stations measuring a 10-minute wind speed value every 3 hours.
As the data was used to represent large scale wind power production, it was upscaled. For this
upscaling, first a 1000 MW wind power production series was produced, so that it would
represent the geographic distribution of a potential wind power production in Finland:
Lapland and Åland archipelago and the Southern coast were reduced to a tenth of large scale
capacity each, and the West coast was given the bulk of wind power production (300 MW in
the South and 400 MW in the North of West coast). Upscaling is presented in Table 1.
21
Lapland
100 MW
Wind speed data
Single turbine data
2-8 turbine data
100
Gulf of Bothnia
50
50
North,
400 MW
200
50
Gulf of Bothnia 50
South,
100
300 MW
50
50
Åland
100 MW
50
50
Gulf of Finland
100 MW
Figure 8. Time series collected for wind power production in Finland.
Table 1. Wind power production data from Finland, and upscaling to 1000 MW wind power.
Site (region)
Kotka (South coast, East)
Loviisa (South coast, East)
Åland archipelago
(South coast, West)
Uusikaupunki (West cst, South)
Eurajoki (West coast, South)
Pori (West coast, South)
Närpiö (West coast, South)
Korsnäs (West coast, South)
Kalajoki, Siikajoki, Hailuoto
(West coast, North)
Lumijoki (West coast, North)
Oulunsalo (West coast, North)
Kuivaniemi (West coast, North)
Lapland
TOTAL
Turbines /wind speed
2 x 1 MW
Hourly wind speed 100 m.a.g.l
12 turbines on 7 sites (80 km):
225 kW Sottunga island, 500 kW
Eckerö, 500 kW Kökar island, 4 x 600
kW Lemland, 2 x 500 kW+600 kW
Finström, 500 kW Vårdö island, 600
kW Föglö island
2 x 1.3 MW
Hourly wind speed 100 m.a.g.l
8 x 1 MW
1 x 750 kW
4 x 200 kW
10 turbines on 4 sites (100 km):
2 x 300 kW Kalajoki, 2 x 300 kW and 2
x 600 kW Siikajoki (2 sites),
2 x 300 kW + 2 x 500 kW Hailuoto
1 x 660 kW
1 x 1.3 MW
6 x 750 kW
8 turbines on 2 sites (100 km):
5 x 600 kW Olos, 2 x 450 kW + 600
kW Lammasoaivi
55 turbines on 21 sites + 2 wind speed
measurement sites
22
Data MW
2.0 MW
2.0 MW
6.32 MW
Upscaled to
50 MW
50 MW
100 MW
2.6 MW
2.0 MW
8.0 MW
0.75 MW
0.8 MW
4 MW
50 MW
50 MW
100 MW
50 MW
50 MW
200 MW
0.66 MW
1.3 MW
3 MW
4.5 MW
50 MW
50 MW
100 MW
100 MW
38 MW
1000 MW
3.3 Data set for Denmark
For Denmark, the system operators Eltra (West DK) and Elkraft System (East DK) have
hourly production data available at their internet sites, starting from year 2000 (Eltra, 2003;
Elkraft, 2003). The maximum distance between the sites in West Denmark is roughly 300 km
North-South and 200 km West-East. For the Eastern part, the dimension is about 200 km
North-South and 100 km West-East. Bornholm island, South of Sweden, is a part of East
Denmark.
Danish data is representing the realised production of thousands of turbines and hundreds of
sites. However, there has been a significant increase in wind power capacity during the two
years: from 1740 MW in start of 2000 to 2524 MW at the end of year 2001 (West Denmark:
1340 MW in the start of 2000, 1790 MW in the start of 2001 and 1970 MW in the start of
2002, East Denmark 390, 503 and 554 MW respectively).
To be correct in converting the hourly production in MWh/h to relative production, as % of
capacity, exact data on each wind farm’s network connection would be needed. This means
making an hourly PTOT time series in formula 1, PTOTi. If the information on capacity addition
(or reduction as some old wind turbines have been taken from operation) was not correct, a
step up in the MW time series at a wrong hour could distort the real production time series.
This would either add more variations or damp the real variations from one hour to the next.
Exact daily data on capacity development in East Denmark was obtained for year 2001. For
West Denmark, and for year 2000 of East Denmark, no exact data on capacity was available.
For these data sets, an approximate hourly MW series has been constructed to convert the data
to % of capacity. For West Denmark, the capacity has been rising at an average rate of 50
kW/hour in 2000 and 13 kW/hour in January 2001, after which a constant capacity has been
used. For East Denmark, the capacity has been rising at an average rate of 16 kW/hour in
2000.
Looking at the capacity increase in Denmark (Elkraft, 2003; Eltra, 2003), it has been quite
linear. The error made here would stay below 50 MW at any hour (the difference between the
approximation used here and the real life). This results in a maximum 3 % error in the hourly
% of capacity values (for example, the production level 84.7 % would in real life be 82.4 % or
vice versa) in the beginning of year 2000 (total capacity 1740 MW) and a maximum 2 % error
at the end of year 2000 (total capacity 2330 MW). The error for the hourly variations are even
smaller, as the capacity increase in practise comes as 1-3 turbines at a time, when the test
operation of a wind farms starts. Assuming a maximum 10 MW instantaneous capacity
increase in an hour, this would be seen as a 0.5 % error in the hourly variation, either
overestimating an upward variation or underestimating a downward variation in the data set
used in this study. There is no detectable error in data in the situation where there is in real
life no increase in the capacity from one hour to the next – an assumed 60 kW increase in
capacity is 0.004 % of the total capacity in the beginning of year 2000 and 0.003 % at the end
of year 2001.
The advantage of the procedure used here is that we get a better knowledge of how much, as
% of capacity, the production has been.
23
3.4 Data set for Sweden
For Sweden, wind power production data was acquired from 2 sites in Southern Sweden
(West and South coast) and 2 sites on the island of Gotland (East coast). From the Northern
part, only one wind speed measurement time series was acquired (SMHI, 2003). From the
Central part of Sweden, no data was available for this study. This is due to late start of hourly
data registration of wind farm production data in Sweden, for most wind farms this was first
started in 2001.
The South Sweden area is about 300 km North-South and 400 km West-East. Taking the one
wind speed measurement available from North of Sweden, this increases the North-South
dimension to 1300 km.
For upscaling, a 1000 MW wind power production series was produced, representing the
geographic distribution of potential wind power production in Sweden. Most of the capacity
was assumed in Southern Sweden with 300 MW West coast, 300 MW South coast and 300
MW Gotland island. 100 MW was assumed in Northern Sweden (1 wind speed series).
3.5 Data set for Norway
For Norway, wind power production data was acquired from one site. However, the data had
missing periods especially for year 2001. Two wind speed measurement time series were
acquired from potential wind power sites in Middle and South Norway, covering parts of
years 1999 and 2000.
Norwegian meteorological institute (NMI) data was well representative for wind power
production: it is measured hourly and with high average wind speeds. 5 sites along the
coastline were used for 2000 and 11 sites for year 2001.
Norway is the largest country when considering the largest dimensions between the potential
wind farm sites: about 1400 km North-South and 700 km West-East. The South Norway area
is about 500 km North-South and 150 km West-East, Middle Norway 300 and 100 km, and
North Norway 400 and 400 km respectively.
For up-scaling, Norway was divided to 3 regions, first aggregating the available data as
simple averages per site for each region South, Middle and North Norway. The total wind
power production was also a simple average: same amount of wind power was assumed to
South, Middle and North Norway.
In Norway data, there were several periods of high wind speeds above the cut out wind speed
of wind turbines (Fig.7). Especially during the first months of 2000 (7.- 8.1., 3.-4.2.,10.-11.2.)
and November, 2001 (3.11. and 10.-11.11.).
24
3.6 Long term yearly production data
Long term data was used to determine the representativeness of the wind resource for the
example years.
Yearly wind production index data was acquired from existing national wind energy statistics
(Laakso, 2003; Carlsted, 2003 and Naturlig energi). Production index data was used because
the yearly wind production data (also available from statistics) needs to be corrected for the
capacity built during the year. As exact average capacity is not known, but only the capacity
in the beginning and end of years, this would result in errors trying to make a representative
and comparable figure for the yearly production. Also, the wind index data reaches farther
back in time than the production data, which has only started in the 90’s for Finland and late
80’s for Sweden and Denmark. Norway wind index data does not yet exist.
Wind power production index is a measure of one year’s production compared with the long
term average production. For Denmark and Sweden, this is derived by looking at the
production of reference turbines, operating since the end of 1980’s. Production index for one
year is the production of those turbines that year divided by the average production of those
turbines over a long reference time period. Index of 100 % means that the production during
the year has been the same as for long term average. For Finland, the production indices are
calculated from the Finnish Meteorological Institute (FMI) wind speed measurements along
the coast, converting the wind speed to power production. In Finland, the coastal areas South
and West experience somewhat different wind resource variations, that is why the production
indices are presented for 4 sites (Laakso, 2003). The long term average period is 15 years,
1987-2001, over which period the wind index is on the average 100 %.
25
4 Large scale wind power production
Examples of the data sets in this study are presented in Fig. 9 for February, 2000. Graphs of
yearly data for years 2000 and 2001, for the 4 countries and their combination, is presented in
Appendix 1.
When studying the effects of wind power on power systems, the wind power data has to
represent large scale wind power production. This means the production of hundreds (or
thousands) of turbines, tens (or hundreds) of sites. Geographical spreading of production
evens out the total production from an area. Duration of calms will be substantially decreased,
as the wind blows almost always at some part of the system area (Giebel, 2001). Maximum
production level will not reach installed nominal capacity, as the wind will not blow as
strongly at all sites simultaneously, and of hundreds or thousands of WTs not all are
technically available at each instant. The extent of the smoothing effect of wind power
production depends mainly on number of sites and distribution of sites over the area, as well
as spatial correlation between the production of the sites (Focken et al, 2001).
In this chapter, a closer look on large scale wind power production is taken, analysing the
patterns of wind power production and how the aggregation of production from a larger area
affects these patterns. The smoothing effect can be seen from most of the statistical analyses
presented in this chapter. The representativeness of the data sets and the example years are
discussed in chapter 5.
26
Nordic, average 39 %
100 %
80 %
60 %
40 %
20 %
0%
1
24 47 70 93 116 139 162 185 208 231 254 277 300 323 346 369 392 415 438 461 484 507 530 553 576 599 622 645 668 691
Denmark, average 40 %
100 %
80 %
60 %
40 %
20 %
0%
1
24 47 70 93 116 139 162 185 208 231 254 277 300 323 346 369 392 415 438 461 484 507 530 553 576 599 622 645 668 691
Finland, average 30 %
100 %
80 %
60 %
40 %
20 %
0%
1
24 47 70 93 116 139 162 185 208 231 254 277 300 323 346 369 392 415 438 461 484 507 530 553 576 599 622 645 668 691
Norw ay, average 46 %
100 %
80 %
60 %
40 %
20 %
0%
1
24 47 70 93 116 139 162 185 208 231 254 277 300 323 346 369 392 415 438 461 484 507 530 553 576 599 622 645 668 691
Sw eden, average 40 %
100 %
80 %
60 %
40 %
20 %
0%
1
24 47 70 93 116 139 162 185 208 231 254 277 300 323 346 369 392 415 438 461 484 507 530 553 576 599 622 645 668 691
Figure 9. Hourly wind power production in February, 2000. The production is as % of installed
capacity (y-axis).The average production during the month is denoted above the curve.
27
4.1 Basic statistics of the wind power production data used
The average production from all seasons in the study period is shown in Table 2. The 12
months are divided into seasons as the following: spring is March, April and May; summer is
June, July and August; autumn is September, October and November; winter is January,
November and December.
First of all, the difference in wind resource is notable: Norway has an excellent wind resource,
with average production of nearly 33 % of capacity compared with 22–24 % for the other
Nordic countries. Denmark has here the lowest production rates, as % of capacity. This is
probably due to the data containing also sites in the inland and sites with older turbines with
20-40 m towers: the rotors are not reaching as good wind resource as the new, 60-100 m high
MW scale turbines. The production in 2000 and 2001 does not yet have large offshore wind
power included, with better wind resource (2 x 160 MW wind farms erected in 2002 and
2003).
It can be seen from Table 2, that year 2000 was considerably more windy than year 2001,
especially winter and spring months. The difference is most pronounced in Denmark and
Sweden and less pronounced in Norway. The production during the summer months is 60–80
% of the yearly average and production during the winter months is 110–140 % of the yearly
average (Table 2).
Table 2. Average wind power production in the Nordic countries in the study period. Wind power
production is presented as relative production, % of installed capacity.
Nordic
Denmark
Finland
Norway
Sweden
Years 2000-2001
25.5 %
22.2 %
23.3 %
32.6 %
23.9 %
Year 2000
27.1 %
24.1 %
24.4 %
33.8 %
26.0 %
Year 2001
24.0 %
20.1 %
22.2 %
31.4 %
21.9 %
Winter 2000
35.8 %
33.5 %
29.2 %
45.0 %
35.5 %
Spring 2000
26.3 %
20.8 %
26.6 %
33.5 %
24.7 %
Summer 2000
17.9 %
17.9 %
16.2 %
21.5 %
16.1 %
Autumn 2000
28.3 %
24.6 %
25.4 %
35.6 %
27.8 %
Winter 2001
28.8 %
21.8 %
23.6 %
41.8 %
27.8 %
Spring 2001
20.5 %
18.8 %
19.7 %
24.6 %
18.9 %
Summer 2001
17.8 %
15.2 %
17.9 %
21.8 %
16.4 %
Autumn 2001
29.5 %
25.0 %
26.5 %
37.7 %
27.7 %
The basic statistics of the yearly time series are presented in Table 3. Wind power production
from the 4 countries and the combination are shown. As a comparison, data from one site is
shown.
28
Table 3. Descriptive statistics of hourly wind power production in the Nordic countries for years
2000-2001. Wind power production is presented as relative production, % of installed capacity. The
width of the areas are presented as largest distance (km) North-South and West-East.
Single
site
Largest distance NS-WE
Denmark
-
Finland
Norway
Sweden
Nordic
300-300 1000-400 1400-700 1300-400 1700-1100
Mean
25.9 %
22.2 %
23.3 %
32.6 %
23.9 %
25.5 %
Median
14.9 %
14.4 %
18.8 %
28.9 %
17.4 %
22.2 %
Standard Deviation
28.2 %
21.4 %
17.9 %
20.1 %
20.6 %
15.5 %
105.0 %
92.7 %
91.1 %
92.8 %
98.5 %
83.6 %
Minimum
0.0 %
0.0 %
0.0 %
0.3 %
0.0 %
1.3 %
Maximum
105.0 %
92.7 %
91.1 %
93.1 %
98.5 %
84.0 %
Largest(24)
102.8 %
89.0 %
84.1 %
89.0 %
94.1 %
80.8 %
0.0 %
0.0 %
0.3 %
0.9 %
0.0 %
1.9 %
Range
Smallest(24)
To take a closer look at the regional wind power production, the same statistics are presented
in Table 4, for the regions of the countries (Denmark and Sweden 2 regions; Norway 3
regions and Finland 5 regions). For the regions, there is clearly not as good smoothing effect
seen, except for the real data for Denmark East and West.
Table 4. Descriptive statistics of hourly wind power production in the different regions of Nordic
countries. Years 2000-2001.
South
Norway
NS-WE
Middle
Norway
North
Norway
500-150 300-100 400-400
DK East
DK West South
North
FI South FI South FI West
coast
coast
Sweden Sweden coast
South
West
East
200-100 300-200 300-400
- 20-50
50-80
250-30
FI West
coast
North
FI
Lapland
150-50
100-100
Mean
31.7 %
37.7 %
28.4 %
20.7 %
22.7 %
23.8 %
28.6 %
22.0 %
25.9 %
23.6 %
23.8 %
21.3 %
Median
26.8 %
28.5 %
22.9 %
12.4 %
14.8 %
16.2 %
15.1 %
12.8 %
17.9 %
17.0 %
17.2 %
14.6 %
StDev
24.5 %
33.1 %
22.7 %
21.7 %
22.0 %
22.2 %
31.4 %
23.9 %
24.7 %
22.0 %
22.1 %
20.8 %
100.0 % 100.0 % 100.0 %
92.3 %
94.0 %
98.6 % 100.0 % 102.1 %
98.7 %
99.4 %
99.4 %
97.5 %
0.0 %
0.0 %
0.0 %
-1.0 %
0.0 %
-0.1 %
-0,1 %
0.0 %
100.0 % 100.0 % 100.0 %
92.3 %
94.0 %
98.5 % 100.0 % 101.1 %
98.7 %
99.3 %
99.3 %
97.5 %
98.9 %
89.9 %
90.8 %
96.6 % 100.0 % 100.8 %
96.7 %
95.4 %
95.4 %
90.0 %
0.0 %
0.0 %
0.0 %
0.0 %
0.0 %
-0,1 %
0.0 %
Range
Minimum
Max
Largest
(24)
Smallest
(24)
0.0 %
0.0 %
99.3 % 100.0 %
0.0 %
0.0 %
-0.1 %
0.0 %
29
0.0 %
0.0 %
-0.4 %
The median is the value in the middle, when sorting all the values in an increasing or
decreasing order. For wind power production it is typical that median is lower than the mean
value. Most of the time, the production is less than average. When aggregating production
from a larger area, the median gets closer to the mean value.
The smoothing effect can be seen in the range of the production, the maximum and minimum
encountered during the years. For the total Nordic time series the production never goes to 0,
however, the lowest production is only 1 % of installed capacity. For one country, the
production can go to 0, even to a slightly negative value, due to consumption of electricity at
the power plants. However, the 24th smallest production value is above 0 (Table 3), so there
are less than 24 hours of 0 production during the two years. The maximum production from
geographically dispersed wind power production stays under 90 % for the Nordic countries.
For a single country, it seldom passes 90 %, as can be seen from the 24th largest value that are
below 90 % for the countries. Sweden is here an exception, which can be explained by the
data being 90 % from the Southern part of Sweden from only 4 sites. Even if we are talking
about large scale wind power production, the production range will still be large compared
with other production forms: maximum production will be 3–4 times the average production,
depending on the area (Table 3; Giebel, 2000).
Another trend of smoothing can be seen in the standard deviation values. Standard deviation
tells about the variability of the hourly time series, it is the average deviation from mean
value:
σ=
nΣx 2 − (Σx) 2
n(n − 1)
(2)
The reduction in variability, measured as the reduction in standard deviation, is depicted in
Fig.10.
East Denmark
Standard deviation of hourly production
Total DK
Finland
100 %
Norway
stdev as % of
single wind farm stdev
120 %
Sweden
80 %
Nordic
60 %
40 %
20 %
0%
0
200
400
600
800
1000
1200
1400
Average diameter of area (km)
Figure 10. Reduction in variability of wind power production: reduction in standard deviation of
hourly time series taken from different areas, as relative to a single site standard deviation (28 % of
capacity). Data from year 2001.
30
For a single turbine, the standard deviation is close to 30 %, somewhat larger than the mean.
For a country, the standard deviation gets closer to 20 %, for a larger country like Norway and
Finland, where the sites are spread 1000 km apart, the standard deviation is less than 20 %.
For the total Nordic time series, the standard deviation is close to 15 % (Table 3). In Fig. 10 it
can be seen that Denmark, Finland and Nordic data represent a reduction, whereas specifically
the Norwegian data series shows far more variations than those of considerably smaller East
Denmark. This was as expected, as the Norwegian data for the areas consists of 2–5 time
series only. For the Nordic data, also a data set where wind power was concentrated in
Denmark (half of the capacity) was made, and there the reduction in standard deviation is to
60 % of the single site value, compared with the nearly to 50 % for an evenly distributed wind
power production (Fig. 10).
4.2 Frequency distributions of wind power production
To take closer look at wind power production, the hourly production of years 2000 and 2001
are plotted as frequency distributions. In Fig.11 the data is grouped with the scale in x axis as
following: 0 means the number of values below or equal to 0 ; 5 % means the number of
values above 0 and below or equal to 5 % etc.
25 %
25 %
Norway, average 33 %
Single wind farm, average 22 %
20 %
frequency, % of time
frequency, % of time
20 %
15 %
10 %
5%
Denmark, average 22 %
15 %
10 %
5%
0%
0%
0%
10 %
20 %
30 %
40 %
50 %
60 %
70 %
80 %
90 %
100 %
0%
10 %
20 %
production, % of capacity
30 %
40 %
50 %
60 %
80 %
90 %
100 %
25 %
25 %
Finland, average 23 %
Nordic, average 26 %
20 %
20 %
frequency, % of time
frequency, % of time
70 %
production, % of capacity
15 %
10 %
Sweden, average 24 %
15 %
10 %
5%
5%
0%
0%
0%
10 %
20 %
30 %
40 %
50 %
60 %
70 %
80 %
90 %
100 %
0%
production, % of capacity
10 %
20 %
30 %
40 %
50 %
60 %
70 %
80 %
90 %
100 %
production, % of capacity
Figure 11. Frequency distribution of wind power production from one site, from a country and for a
theoretical total Nordic production. Example years 2000 and 2001. The production values at x axis
note the upper value of the range.
31
It can be seen in Fig.11, that large scale production of wind power means shifting the most
frequent ranges to the middle of the graph. For the single site, the production is almost half of
the time below 10 % of capacity. For the wind power scatter to all Nordic countries, the
production is most of the time in between 5...30 % of capacity, and is seldom below 5 % or
above 70 % of capacity.
production (% of capacity)
The probability of wind power production can also be presented as a duration curve. The
duration curve of power production is often used in the energy sector to illustrate the time the
power plant produces a certain power level. In Fig.12, the Nordic wind power production for
year 2000 is shown chronologically (the varying curve) and as a duration curve, where the
production values are sorted in descending order before drawing the curve. The duration
curve does not tell about the correlation between consecutive values, for this a persistence
study is made separately in chapter 4.5.
100 %
2000
80 %
60 %
40 %
20 %
0%
1
741
1481
2221
2961
3701
4441
5181
5921
6661
7401
8141
hour
Figure 12 Example of data for this study: the total Nordic wind power production, as a chronological
time series and as a duration curve.
100 %
Nordic, average 26 %
production % of capacity
90 %
80 %
Denmark, average 22 %
70 %
Single turbine, average 26 %
60 %
50 %
40 %
30 %
20 %
10 %
0%
1
741 1481 2221 2961 3701 4441 5181 5921 6661 7401 8141
hour
Figure 13. The effect of geographical spreading is to flatten the duration curve of wind power
production. Example of year 2000 hourly data, where wind energy distributed to all 4 Nordic
countries is compared with one of the wind farms and one of the countries (Denmark). Average
production for the curves is denoted in the legend text.
32
In Fig. 13, the smoothing effect is presented as duration curves. The duration curves for the
countries are presented in Fig. 14. Here again it can be seen that the Norwegian wind power
production is at a higher level than for the other countries, and also the smoothing effect is
stronger. The Danish wind power production shows less smoothing effect than the data sets
for the other countries. This is due to Denmark being far smaller area than the other countries.
100 %
Norway, average 33 %
production % of capacity
90 %
Sweden, average 24 %
Denmark, average 22 %
Finland, average 23 %
80 %
70 %
60 %
50 %
40 %
30 %
20 %
10 %
0%
1
741 1481 2221 2961 3701 4441 5181 5921 6661 7401 8141
hour
Figure 14. Duration curves for the wind power production in the 4 Nordic countries, year 2000 data.
Average production is denoted in the legend text.
4.3 Seasonal variation of wind power production
In Central and Northern Europe, there is a distinct seasonal variation in wind power
production: more production in winter than in summer. For example in the Nordic countries,
60–70% of yearly production comes during 6 winter months (Fig.15). The production during
the winter months is 110–140 % of the yearly average and production during the summer
months 60–80 % of the yearly average (Table 2). This is also reflected in the range of
production values, for example, the hourly data for Nordic countries for these example years
2000 and 2001 ranges between 1...61 % in the summer and 2...85 % in the winter.
Frequency distributions for the 4 seasons are presented in Fig.16. Duration curves for summer
and winter are presented in Fig.17 for Denmark and the combined Nordic wind power
production.
33
Electric
consumption
2002
Average
production
1995-2002
1995
% of yearly production /consumption
15 %
10 %
1996
1997
1998
5%
1999
2000
0%
2001
1
2
3
4
5
6
7
8
9
10
11
12
2002
Month
Figure 15. Seasonal variation of total wind power production in Finland in 1997-2001. Average of
1992-2001 is shown (line) together with the electric consumption (dotted line).
Nordic spring, average 24 %
30 %
Finland spring, average 23 %
30 %
Finland summer, average 17 %
Nordic summer, average 18 %
Nordic autumn, average 29 %
frequency (% of time)
frequency (% of time)
25 %
Nordic winter, average 32 %
20 %
Nordic 2000-2001, average 26 %
15 %
10 %
25 %
Finland autumn, average 26 %
20 %
Finland winter, average 26 %
Finland 2000-2001, average 23 %
15 %
10 %
5%
5%
0%
0%
frequency (% of time)
frequency (% of time)
100 %
90 %
80 %
Denmark winter, average 28 %
Denmark 2000-2001, average 22 %
15 %
70 %
Norway summer, average 22 %
25 %
Denmark autumn, average 25 %
20 %
60 %
Norway spring, average 29 %
30 %
Denmark summer, average 17 %
25 %
50 %
40 %
production level (% of capacity)
Denmark spring, average 20 %
30 %
30 %
20 %
10 %
0%
100 %
90 %
80 %
70 %
60 %
50 %
40 %
30 %
20 %
10 %
0%
production level (% of capacity)
10 %
Norway autumn, average 37 %
Norway winter, average 43 %
20 %
Norway 2000-2001, average 33 %
15 %
10 %
5%
5%
0%
0%
100 %
90 %
80 %
70 %
60 %
50 %
40 %
30 %
20 %
10 %
0%
100 %
90 %
80 %
70 %
60 %
50 %
40 %
30 %
20 %
10 %
0%
production level (% of capacity)
production level (% of capacity)
Figure 16. Difference in frequency distributions of wind power production for seasons: the lower
production rates have a higher probability in summer, and higher production rates are more probable
in winter.Average production during the seasons is denoted in the legend text.
34
Denmark, summer, average 17 %
production (% of capacity)
100 %
90 %
Nordic, summer, average 18 %
80 %
Denmark, winter, average 28 %
70 %
Nordic, winter, average 32 %
60 %
50 %
40 %
30 %
20 %
10 %
0%
1
741
1481
2221
2961
3701
hour
Figure 17. The wind power production is higher during the 3 winter months (upper curves: January,
February and December) than the 3 summer months (lower curves: June, July, August). Duration
curves for production in 2000 and 2001.
4.4 Diurnal variation of wind power production
Wind is driven by weather fronts and a daily pattern caused by the sun, so depending whether
one of these dominates there is either significant or hardly any diurnal pattern in the
production. Diurnal variation can also be due to local phenomena, for example in California
passes there are morning and evening peaks when wind blows to and from the desert and the
sea. In Europe, there is a tendency for winds starting to blow in the morning and calming
down in the evening (Ireland: Hurley and Watson, 2002; Germany: Ensslin et al, 2000). In
Northern Europe this is mostly pronounced during the summer (Fig.18).
In winter there is not a clear diurnal variation to be seen, except for slightly in Denmark (the
uppermost curves in Fig. 18 graphs). In summer, the average production at 11…18 hours is on
the average above 20 % of capacity compared with less than 15 % of capacity during the
night. Wind power production in Denmark and Sweden experience a more pronounced
diurnal variation, whereas the sites in the northern part of Finland, Sweden and Norway do
not experience any detectable diurnal variation.
The diurnal variation here is presented in Central Europe time, as is used in Denmark,
Norway and Sweden. The hours have a shift for summer time in the spring and back to
normal time in the autumn (Fig 18). For the single sites with whole year data, the hours are in
normal time. A shift in the peak can be seen for the single site data in Fig.19, where data for
all countries are in the same graph. The sun rises from the East, warming up Finland first
(peak at 10-12 Central Europe time, 11-13 Finnish time), Sweden next (peak at 11-13),
Denmark (peak at 13-15) and lastly Norway (peak at 14-17). For Norwegian data, the
smoothing made to single wind speed series to represent wind farm data, makes the peak shift
somewhat to later than it should be (2-hour-sliding average of wind speeds was done).
35
35 %
30 %
All data Denmark,
average 22 %
25 %
Spring Denmark,
average 20 %
20 %
30 %
Summer Denmark,
average 17 %
15 %
10 %
Autumn Denmark,
average 25 %
5%
Winter Denmark,
average 28 %
production (% of capacity)
production (% of capacity)
35 %
All data Finland,
average 23 %
25 %
Spring Finland,
average 23 %
20 %
Summer Finland,
average 17 %
15 %
10 %
Autumn Finland,
average 26 %
5%
Winter Finland,
average 26 %
0%
0%
0 2 4 6 8 10 12 14 16 18 20 22
0 2 4 6 8 10 12 14 16 18 20 22
hour of day
hour of day
45 %
35 %
All data Norway,
average 33 %
35 %
Spring Norway,
average 29 %
30 %
Summer Norway,
average 22 %
25 %
20 %
Autumn Norway,
average 37 %
15 %
Winter Norway,
average 43 %
30 %
production (% of capacity)
production (% of capacity)
40 %
10 %
All data Sweden,
average 24 %
25 %
Spring Sweden,
average 22 %
20 %
Summer Sweden,
average 16 %
15 %
10 %
Autumn Sweden,
average 28 %
5%
Winter Sweden,
average 32 %
0%
0 2 4 6 8 10 12 14 16 18 20 22
0 2 4 6 8 10 12 14 16 18 20 22
hour of day
hour of day
Figure 18. For the Nordic countries, diurnal variation is more pronounced in summer time
Single sites
35 %
0,35
All data Nordic,
average 26 %
25 %
Spring Nordic,
average 26 %
20 %
Summer Nordic,
average 18 %
15 %
10 %
Autumn Nordic,
average 26 %
5%
Winter Nordic,
average 26 %
0%
0,3
production (% capacity)
production (% of capacity)
30 %
0,25
0,2
0,15
North Norway
South Norway
0,1
South Sweden
North Sweden
0,05
West Denmark
South Finland
0
0 2 4 6 8 10 12 14 16 18 20 22
0
hour of day
2
4
6
8
10
12
14
16
18
20
22
hour of day
Figure 19. Diurnal variation of wind power production for some example sites. For North Norway,
Sweden and Finland, the diurnal variation is practically non existent also for summertime.
36
4.5 Persistence of wind power production
Frequency distributions and duration curves give some idea of how often certain production
levels occur. However, for a varying power production like wind power, also persistence in
different production levels is of interest – how long does a certain production level last?
There are two special cases, presenting the greatest challenges in integration of wind power in
the system: duration of calms or low wind power production, as well as occurrence of the
peaks, which are specially pronounced in wind power production. This analysis gives also
insight into how the example years 2000 and 2001 differ in this respect.
4.5.1 Duration of calms
Duration of calms has here been defined as time when wind power production is less than 1 %
of capacity. As the average production is of the order of 20–25 % of capacity, this can also be
put as about 4–5 % of average production. Additionally low production persistence has been
studied: when wind power production is less than 5 % of capacity (roughly 20 % of average
production). Production level of 10 % of capacity is already almost half of average
production, and wind power production is almost a third of the time below 10 % level (for the
total Nordic production, almost 15 % of time, Fig. 11). That is why it is not considered as a
calm period.
14
6
31
34
29
27
25
23
21
1
29
27
25
23
21
19
17
15
13
11
9
7
5
3
1
0
19
2
17
4
15
6
13
8
2000, total 134 hours
2001, total 144 hours
11
10
Sweden
9
number of periods
2001, total 11 hours
20
18
16
14
12
10
8
6
4
2
0
7
2000, total 22 hours
5
14
number of periods
22
duration of calm, production less than 1 % of cap (hours)
duration of calm, production less than 1 % of cap (hours)
Norway
19
1
29
27
25
23
21
19
17
15
13
9
11
0
7
2
0
5
2
16
4
13
4
8
10
6
7
8
12
2000, total 404 hours
2001, total 430 hours
10
4
number of periods
10
3
Denmark
12
2001, total 88 hours
1
number of periods
12
28
2000, total 92 hours
3
Finland
25
14
duration of calm, production less than 1 % of cap (hours)
duration of calm, production less than 1 % of cap (hours)
Figure 20. Duration of calms for wind power production, number of different length periods when
production below 1 % of capacity. The graphs for countries do not all have the same scale.
37
In Denmark, the production was below 1 % of capacity a total of nearly 5 % of time, where as
for the larger areas of Finland and Sweden, this was about 1 % of time. For Norway, the
calms were very rare (0.2 % of time). The longest duration of calm (production below 1 % of
capacity) was 35 hours for Denmark in 2000. For Sweden, it was 18 hours, for Finland 14
hours and for Norway 5 hours. In Fig.20 it can be seen that for Norway, the total amount of
hours below 1 % of capacity is half in 2001 compared to 2000, this can be explained by more
data series for year 2001. For a total Nordic data set, there were no calms, the production is
always above 1 % of capacity. The production was below 5 % of capacity about 2 % of time
(Fig.21).
24
22
20
18
16
14
12
10
8
6
4
2
0
2000, total 200 hours
35
33
31
29
27
25
23
21
19
17
15
13
11
9
7
5
3
2001, total 216 hours
1
number of periods
There are not significantly more and longer calms in year 2001 data than in year 2000 data.
This is somewhat surprising: as year 2001 was a lower wind year than year 2000 (Table 2), so
intuitively also the calm periods could have been more.
duration of low production, less than 5 % of cap (hours)
Figure 21.Duration of low production in a total Nordic wind power production, number of different
length periods when production less than 5 % of capacity.
4.5.2 Peaks of wind power production
Peak production has here been studied for the level of above 75 % of capacity. As the average
production is of the order of 25 % of capacity, this can also be defined as roughly three times
the average production.
In 2000, there was one long period with high wind power production exceeding 75 % of
capacity: 37 hours in Finland, 34 hours in Sweden and Denmark and 27 hours in Norway. In
addition, there were 1-3 periods of about 20 hours long high production. For the Nordic data,
peaks of more than 75 % are rare, 84 hours in 2000 and 34 hours in 2001, with maximum
duration of 14 hours.
In all the countries except Norway, there are more and longer periods of peak power
production in year 2000 data than in year 2001 data. This was as expected, as year 2000 saw a
better wind resource (Table 2) so also the high wind periods are supposedly more.
38
Duration of peaks
Duration of peaks
14
6
Duration of peaks
Sweden
40
37
34
40
37
34
31
28
25
22
19
16
13
7
2001, total 166 hours
1
duration of peak, production more than 75 % of cap (hours)
2000, total 396 hours
10
number of periods
29
27
25
23
21
19
17
15
13
11
9
7
5
24
22
20
18
16
14
12
10
8
6
4
2
0
4
2000, total 211 hours
2001, total 295 hours
3
31
Duration of peaks
Norway
1
number of periods
25
duration of peak, production more than 75 % of cap (hours)
duration of peak, production more than 75 % of cap (hours)
24
22
20
18
16
14
12
10
8
6
4
2
0
22
1
40
37
34
31
28
25
22
19
16
7
13
0
10
2
0
19
4
2
16
4
13
6
8
7
8
2001, total 223 hours
10
10
number of periods
10
4
2000, total 372 hours
12
2001, total 33 hours
1
number of periods
Denmark
2000, total 91 hours
4
Finland
12
28
14
duration of peak, production more than 75 % of cap (hours)
Figure 22. Length of high wind power production periods in example years 2000 and 2001.
4.6 Correlation of wind power production
Cross-correlation (rx,y) is a measure of how well two time series follow each other. It is near
the maximum value 1 if the ups and downs of the production occur simultaneously, near the
minimum value –1 if there is a tendency of decreasing production from one site when
increasing production at the other site, and close to zero if the two are uncorrelated, and the
ups and downs of production do not follow each other at two sites.
rx , y =
1 n
∑ ( xi − µ x )( yi − µ y )
n i =1
σ xσ y
,
(3)
where µ denotes the average, σ the standard deviation and n the number of points in the time
series.
Correlation can also be calculated for a single time series but with time lags. This is called
autocorrelation. For wind power production, the autocorrelation decreases soon with
increasing time lag, already at 12 hour lag the correlation becomes weak (Pryor & Barthelmie,
2001)
If wind production data is not correlated, there can be strong winds in one place at the same
time as weak winds in the other. When distributing wind power production to a larger area,
39
the total production will be smoother and less variable, if the correlation between the sites is
low.
The cross-correlations were calculated for all sites in the Nordic countries for one year, 2001,
when the data available included most sites, altogether 33 time series. Some of the time series
were aggregated production data from a larger area, for which the coordinates were estimated
from the centre of the area. The results are presented in Fig. 23. The cross-correlation
decreases fast at first, rxy =0.7 for distance of about 100 km and 0.5 for distance of about 300
km, after which the decrease is slower.
There is significant variation in the cross correlation coefficients for a similar distance, as is
expected. The correlation becomes weak, below 0.5, with distances above 200-500 km. When
local phenomena influence the wind resource, the winds do not correlate with sites even some
200 km apart. In Fig. 23, the lowest cross-correlations are slightly negative, for Finnish
Lapland with Southern Norway sites. For the westernmost site in Southern Norway, the
correlation is weak for all other sites, the lowest points in Fig. 23 for distances of 200...800
km come from there. Slightly negative correlations between two points in Europe have been
reported from weather data from Ireland/Portugal (1500 km apart) and Spain/Greece (3000
km apart) (Giebel, 2001). The results from correlation between weather station wind speed
based data calculated from 9 years in Finland are similar to the ones here for year 2001
(Tammelin&Nurmi, 2001). There is not a significant change in correlation coefficients
calculated from different years. A year of hourly data contains enough different weather
situations to be able to determine the correlation between the wind power production at
different sites.
The cross-correlation can be modelled by exponential fitting, decay parameters (D) of
500…700 have been reported (Giebel, 2001). For this data, D= 500 fits the data (Fig. 23).
1
0,9
correlation coefficient
0,8
0,7
y=exp(-d/500)
0,6
0,5
0,4
0,3
0,2
0,1
0
-0,1
0
200
400
600
800
1000 1200 1400 1600 1800 2000
distance between sites, d (km)
Figure 23. Cross correlation coefficients for the sites in the Nordic data for year 2001.
Looking at large scale wind power production in the countries and regions, the correlations
are calculated for 2 years of data (2000-2001) and presented in tables 5-6. For the four
40
countries, Swedish and Danish wind power production is correlated (with the assumption here
that most of the Swedish wind power is in the Southern part of Sweden). Wind power
production in the other countries is only weakly correlated, with lowest correlation between
Denmark and Finland.
Table 5. Cross correlation coefficients between wind power production in the Nordic countries.
Norway
Denmark
Sweden
Norway
1.00
Denmark
0.37
1.00
Sweden
0.41
0.77
1.00
Finland
0.44
0.33
0.45
Finland
1.00
Taking a closer look at the regions in the Nordic countries, the largest correlation is again for
wind power production in West and East Denmark and South Sweden. These are the areas
with least distance apart. Also the two areas in Southern part of Finland are strongly
correlated. For other areas, the correlation is not strong. There is weak correlation (0.4...0.5)
between the areas in Lapland (the Northern part of Norway, Sweden and Finland), between
Southern Norway and Denmark/South Sweden, between South Sweden and the Southern
areas of Finland, between the northernmost West coast of Finland and Lapland, Southern and
Western parts of Finland, and between Middle and North Norway. There is practically no
correlation between Lapland (North Norway, Sweden and Finland) and the Southern areas
(Denmark, South Sweden, Norway and to some extent South Finland).
Table 6. Cross correlation coefficients between the regional wind power production in the Nordic
countries.
NO
NO
NO
South Middle North
DK
DK
SE
SE
East
West South North
Elkraft (Eltra)
FI
South
coast
East
FI
South
coast
West
FI
West
coast
South
FI
West
coast
North
NO S
1.00
NO M
0.33
1.00
NO N
0.21
0.41
1.00
DK E
0.37
0.16
0.15
1.00
DK W
0.48
0.21
0.16
0.86
1.00
SE S
0.40
0.26
0.20
0.78
0.75
1.00
SE N
0.12
0.32
0.41
0.13
0.15
0.17
1.00
FI SE
0.22
0.24
0.16
0.26
0.26
0.38
0.26
1.00
FI SW
0.26
0.33
0.23
0.31
0.32
0.50
0.28
0.52
1.00
FI WS
0.28
0.33
0.28
0.30
0.33
0.42
0.32
0.52
0.71
1.00
FI WN
0.18
0.23
0.37
0.22
0.21
0.25
0.31
0.33
0.33
0.57
1.00
FI Lappi
0.07
0.11
0.37
0.08
0.06
0.07
0.39
0.13
0.13
0.23
0.41
41
FI
Lapland
1.00
4.7 Short term variations of wind power production
For power system operation, the variations from day to day, hour to hour and minute to
minute are of interest. The larger the area, the longer time scales are affected by smoothing
effect. Inside a WF, all the WTs will experience different gusts (seconds), but the hourly wind
power production will see approximately the same ups and downs. In a larger area covering
several hundreds of km, the weather fronts causing high winds will not pass simultaneously
but the good and poor months will occur same time. This can be seen in Fig. 24, where the
decreasing correlation of the variations is depicted for different time scales (Ernst, 1999). The
correlation is here calculated for the differences between consecutive production values (∆P).
For the time series of production values (P), the correlation does not decrease as rapidly as
shown here (Fig.23).
Correlation coefficient for the variations
1
0.8
0.6
12h average
4h average
0.4
2h average
0.2
1h average
30min average
5min average
0
0
100
200
300
400
500
600
Distance (km)
Figure 24. Variations will smooth out faster when the time scale is small. Correlation of variations for
different time scales, example from Germany. (Source: B.Ernst,1999)
4.7.1 The in-hour variations
Already the inertia of large rotating blades of a wind turbine will smooth out the very fast
gusts of wind. For variable speed wind turbines, the second-to-second variations will be
absorbed in the varying speed of the rotor. For a wind farm, the second-to-second variations
will smoothed out, as the same gusts will not occur simultaneously at all turbines, situated
several hundred meters apart.
The extreme ramp rates recorded from one 103 MW wind farm are 4…7 % of capacity in a
second, 10…14 % of capacity in a minute and 50…60 % of capacity in an hour (Parsons et al,
2001). These examples are from a limited area compared with system operation: large wind
42
farm or 3 smaller wind farms some 10 km apart. For a larger area of geographically dispersed
WFs, the second and minute variations will not be significant.
For the 15 min variations in Denmark, the production can vary 8.4 % of capacity 6 times per
month, and the maximum is 11 % (Nordel, 2000). This is not as much as for the hourly
variations, as seen in the following chapter.
There are means to reduce the fast variations of wind power production. Staggered starts and
stops from full power as well as reduced (positive) ramp rates could reduce the most extreme
fluctuations, in magnitude and frequency, over short time scales (Kristoffersson et al, 2002).
This is at the expense of production losses, so any frequent use of these options should be
weighed against other measures (in other production units) in cost effectiveness.
4.7.2 The hourly variations
The hourly variation is here defined as the power difference between two consecutive hours.
It is here measured relative to the nominal capacity, to compare it with several countries with
different amounts of capacity installed.
hourly variation (% of cap.)
hourly variation (% of cap)
∆Pi = Pi − Pi −1 ; ∆pi = pi − pi −1
(4)
15 %
2000
10 %
5%
0%
-5 %
-10 %
-15 %
15 %
2001
10 %
5%
0%
-5 %
-10 %
-15 %
1
741
1481
2221
2961
3701
4441
5181
5921
6661
7401
8141
hour
Figure 25. Hourly variations from Nordic wind power production, chronological time series and
duration curve, years 2000 and 2001.
43
For large scale, dispersed wind power production there will be a significant smoothing effect
in the hourly variations. The correlation of the variations between two WTs decreases faster
than the correlation of the production. For hourly variations, the correlation becomes weak
already in distances less than 100 km (Fig.24, Ernst, 1999). Correlation of hourly variations
for the countries and regions were calculated, and most of them were between –0.01 and 0.04,
so there is no correlation between the hourly variations. Hourly variations in East and West
Denmark are weakly correlated (0.46). For the other closest regions, South Sweden/Denmark,
South Norway/West Denmark as well as the Western part of Finland, the correlation of
variations is below 0.2.
Nordic
hourly variation (% cap)
25 %
20 %
15 %
10 %
5%
0%
-5 %
-10 %
-15 %
-20 %
-25 %
hourly variation (% cap)
In Figs 25 and 26 the amount of hourly variations are shown as duration curves. From the
hourly time series of wind power production, the hourly variation as the difference between
the production at consecutive hours, and these values have been sorted in descending order. In
the figure, 0 % means that the power production keeps on the same level and does not vary
from one hour to another, positive values indicate situations when wind power production is
increasing, and negative values for decreasing production.
25 %
20 %
15 %
10 %
5%
0%
-5 %
-10 %
-15 %
-20 %
-25 %
Nordic
Norway
Finland
Denmark
Sweden
1
741
1481
2221 2961 3701 4441 5181
5921 6661 7401 8141
hour
Figure 26. Variation of wind power production from one hour to the next. Duration curve of
variations, as % of installed capacity, for the Nordic countries, year 2001.
44
In Appendix 2, the hourly variations of wind power production for years 2000 and 2001 are
shown for the 4 countries.
Largest hourly variation is about ± 30 % of capacity when the area is in the order of 200x200
km2 (like West/East Denmark), ± 20 % of capacity when the area is in the order of 400x400
km2 (like Germany, Denmark, Finland, Iowa USA) and about ± 10 % in larger area covering
several countries, like the four Nordic countries (ISET, 2002; Holttinen, 2002; Milligan &
Factor, 2000). For this Nordic data, largest hourly variations are 11 % up and 10 % down. For
Norway and Sweden, despite the large area, the variations are higher than for Denmark and
Finland. This is due to limited number of sites in the data sets. The Nordic variations are
probably overestimated due to this.
These are the extreme values, for most of the time the hourly variations will stay inside ± 5 %
of installed capacity (Fig. 26 and Table 7). It is notable, that as the average production is
about 25 % of capacity, this 5 % of capacity represents 20 % of average power. For the
countries, the hourly variations are more than 5 % of capacity 6…20 % of time. For Denmark
this is 10 % of time, so probably the large variations of Norway and Sweden data sets are due
to too few time series in the countries to represent the variations right. Omitting Norway and
Sweden, the conclusion is that the hourly variations of large scale wind power production are
about 90 % of time between ± 5 % of capacity and 99 % of time between ± 10 % of capacity.
For the total Nordic time series the hourly variations are about 98 % of time between ± 5 % of
capacity (Table 7).
Table 7. Largest hourly variations in the wind power production for Nordic countries, years 2000 and
2001.Maximum variations are as % of installed capacity. The portion of time that the variations are
more than 5 or 10 % of capacity is also presented.
Nordic, evenly
Denmark
Finland
Sweden
Norway
2000
2001
2000
2001
2000
2001
2000
2001
2000
2001
10.9 %
11.4 %
20.1 %
16.7 %
14.0 %
16.2 %
33.5 %
34.5 %
26.9 %
24.8 %
max downvariation
-12.2 %
-10.1 %
-23.1 %
-18.0 %
-15.5 %
-14.9 %
-31.1 %
-31.6 %
-21.2 %
-19.6 %
above 5 %
1.1 %
0.8 %
4.9 %
4.0 %
3.0 %
3.2 %
10.0 %
8.8 %
10.1 %
6.2 %
below -5 %
1.0 %
0.7 %
4.9 %
3.7 %
3.2 %
3.0 %
9.6 %
8.2 %
10.5 %
5.7 %
above 10 %
0.0 %
0.0 %
0.6 %
0.5 %
0.2 %
0.2 %
2.2 %
1.9 %
1.9 %
0.7 %
below -10 %
0.0 %
0.0 %
0.5 %
0.4 %
0.2 %
0.2 %
1.8 %
1.7 %
1.7 %
0.6 %
max upvariation
The difference between year 2000 and 2001 is not very pronounced. For Norway, the
difference is the largest, which is probably due to better data set for year 2001 (more sites).
The largest variation in Denmark was Tuesday evening 8.2.2000 at 21-22 hours up and
Sunday afternoon 30.1.2000 at 15-16 hours down. For the Nordic data set largest up-variation
was 15.11.2001, and surprisingly during the night, at 01-02 hours. This was due to wind
power increasing in all countries simultaneously, but for Norway, the increase was 25 % of
capacity, which is probably overestimated. The largest down-variation was Wednesday
45
15.3.2001, at 15-16 hours, when there was a large variation (20 %) in Denmark
simultaneously with nearly 20 % variation in Norway and more than 10 % variation in
Sweden. If more representative data for Norway was available, the largest Nordic variations
would probably stay in ±10 % of capacity.
Probability of significant variations is a function of production level. Significant changes
occur most probably when wind farms are operating between 25...75 % of capacity, as this is
the steep part of the power curve when changes in wind speed produce largest changes in
power output of the turbines (Poore & Randall, 2001). For large scale wind power, the
production is rarely above 75 %, so an analysis to Nordic data was done for the production
level of above 20 % of capacity (at the first hour). Hourly variations were analysed for these
periods only. Example from duration curve of variations is shown in Fig. 27. It can be seen,
when comparing for all the hourly variations in Fig. 26, that the large variations occur nearly
twice as often for the countries when looking this way (Table 8). For the total Nordic data set
the difference is not that large.
Table 8. Largest hourly variations in the wind power production for Nordic countries, when taking
only the periods with production more than 20 % of capacity (years 2000 and 2001).
Nordic
Denmark
Finland
Norway
Sweden
time above 20 %
56.09 %
40.40 %
47.64 %
66.81 %
44.71 %
max up-variation
11.4 %
20.1 %
16.2 %
26.9 %
33.5 %
-12.2 %
-23.1 %
-15.5 %
-21.2 %
-31.6 %
time above 5 %
1.5 %
9.0 %
5.1 %
10.1 %
14.6 %
time below - 5 %
1.6 %
9.7 %
6.0 %
11.3 %
17.1 %
time above 10 %
0.0 %
1.2 %
0.4 %
1.7 %
3.7 %
time below - 10 %
0.0 %
1.2 %
0.3 %
1.7 %
3.8 %
max down-variation
hourly variation (% of cap
Denmark, -23 %…+20 %
25 %
20 %
15 %
10 %
5%
0%
-5 %
-10 %
-15 %
-20 %
-25 %
1
741 1481 2221 2961 3701 4441 5181 5921 6661
hour
Figure 27. Duration curve of hourly variations, when the initial production level has been above 20 %
of capacity. Data from years 2000 and 2001. (The x scale is not the same as in Fig.26, as the total
amount of hours with this production level is not as much as the total number of hours in 2001.)
46
Reductions in standard deviation for hourly time series were presented in chapter 4.1 before,
this is a measure of reduced variability in the time series with geographic dispersion of wind
power. The standard deviation of hourly time series will reduce from 25–30% for a single site
to 18–22 % for a larger area, that is to 70–80 % of the single site value (Fig.10 and Focken et
al, 2001). For the Nordic area, the reduction is to almost half of the one site value (σ = 15 %).
The standard deviation of the time series of fluctuations ∆P will decrease even faster, from
about 10 % for a single turbine to less than a third (3 %) for an area like West Denmark
(Milborrow, 2001). For Nordic data, the reduction in maximum variations and standard
deviation of variations is presented in Fig.28. The Norway and Sweden data give again larger
standard deviation values than Denmark and Sweden, due to lack of real large scale wind
power data.
Standard deviation of hourly variations
Maximum hourly variation
80 %
60 %
40 %
20 %
0%
-20 %
-40 %
-60 %
-80 %
Finland
Norway
Sweden
Nordic
1100
1000
900
800
700
600
500
400
300
200
1100
1000
900
800
700
600
500
400
300
200
100
0
100
-100 %
East Denmark
Total DK
10 %
9%
8%
7%
6%
5%
4%
3%
2%
1%
0%
0
stdev as % of single wind farm stdev
maximum hourly variation (% cap)
100 %
Average diameter of area (km)
Average diameter of area (km)
Figure 28. Maximum hourly variation in wind power production for the data for Nordic countries.
As can be seen from Fig. 28, the smoothing effect is more pronounced with more turbines and
more separation. The smoothing effect of a specified area has a limit, that is, the time series
will not get smoother if more and more turbines are added from the same area. For Germany,
for example, it has been estimated that 30 sites will be enough to get the low variations
(Focken et al, 2001). After saturation, the only way to increase the smoothening will be to
increase the area – which has a limit somewhere, too. In Fig. 28 it is also obvious that
increasing the area from that of Denmark, the decrease in the statistical parameters shown
here is slower.
Diurnal variations in output can help indicate when significant changes in output are most
likely to occur (Poore & Randall, 2001). The average hourly variations of wind power
production are zero – there are as much up and down variations. However, when plotting the
average hourly variation as of time of day, the average is no longer zero for all hours of the
day. There are more upward changes during the morning hours and more downward changes
during the afternoon hours, as can be seen in Fig. 29. This is more pronounced during
summer, as is the diurnal variation of the production (chapter 4.4). Also the maximum
variations in the data set occur in morning hours for the upward changes and in the evening
hours for the downward changes. The maximum variations are less in summer.
47
Nordic
4%
Denmark
average hourly variation (% cap
average hourly variation (% cap
3%
Sweden
2%
Nordic
4%
Norway
Finland
1%
0%
-1 %
-2 %
-3 %
-4 %
Summer 2000
Norway
Denmark
Sweden
Finland
3%
2%
1%
0%
-1 %
-2 %
-3 %
-4 %
22
20
18
16
14
12
10
8
6
4
2
0
22
20
18
16
14
12
10
8
6
4
2
0
hour of day
hour of day
20 %
20 %
15 %
15 %
10 %
10 %
5%
5%
0%
0%
-5 %
-5 %
-10 %
-10 %
-15 %
-15 %
-20 %
-20 %
Summer 2000
22
20
18
16
14
12
10
8
6
4
2
0
22
20
18
16
14
12
10
8
6
4
2
0
Figure 29. Diurnal dependence of variations. All data and summer 2000. Above: average hourly
variations, as of time of day. Below: maximum variations, up (positive) and down (negative).
4.7.3 Variations for longer time scales
For longer time scales, 4…12 h variations, short term prediction tools for wind power give
valuable information on the foreseeable production levels, and expected variations of wind
power production.
From the Nordic data set, the maximum 4- and 12-hour-variations are presented in table 9.
The range of 4 hour variations is about ± 50 % of capacity for one country. This has also been
reported for a longer following period from Germany (ISET, 2002). For the Nordic area it is ±
35 % of capacity according to this 2-year data set.
The maximum 12-hour variation for the Nordic area is ± 50 % of capacity. Taking larger
areas, like the Northern Europe, and more years of data, ± 30 % change in production 12
hours ahead occurs about once a year (Giebel, 2000).
48
Table 9. Maximum variations from the Nordic wind power production (hourly values from years 2000
and 2001).
Nordic
4 hour variations: max down
4 hour variations: max up
12 hour variations: max down
12 hour variations: max up
Denmark
Finland
Norway
Sweden
-31.1 %
-61.9 %
-35.9 %
-45.2 %
-56.6 %
34.1 %
52.9 %
51.6 %
55.2 %
65.1 %
-45.0 %
-73.6 %
-66.6 %
-84.8 %
-69.1 %
52.7 %
79.1 %
72.9 %
74.2 %
82.5 %
4.8 Predictability of wind power production
Wind power prediction plays an important part in the system integration of large scale wind
power. When the share of installed wind power is significant, the knowledge of the on-line
production and predictions 1…36 hours ahead are needed. Day-ahead predictions help the
scheduling of conventional units: planning the start-ups and shut-downs of slow starting units
in an optimised way, keeping the units running at best possible efficiency, saves fuel and thus
operational costs of the power plants. Predictions 1-2 hours ahead help keeping up the optimal
amount of regulating capacity at the system operators’ use (Milligan et al, 1995). In wind
power production forecasts, as is the case for load forecasts, too, the errors decrease when
forecasting for a larger area (Holttinen et al, 2002).
Predictability is most important at times of high wind power production, and up to 6 hours
ahead, giving enough time to react on varying production also by start-ups and shut-downs of
most of the thermal power plants. An estimate of the uncertainty, especially the worst case
error is also relevant information.
Forecast tools for wind power production are still under development and improvements are
expected (Giebel et al, 2003). The predictions of the power production 12 hours-ahead or
more rely almost entirely on meteorological forecasts for local wind speeds. In northern
European latitudes, the variations of wind power production occur due to meteorological
weather systems passing the area, causing high winds, which calm down again. The largest
error component comes from the wind speed forecast of the Numerical Weather Prediction
models. So far the accuracy of ± 2-3 m/s, ± 3-4 hours has been enough for wind speeds in
weather forecasts, but electricity market (and system) requires more precise knowledge of
wind power production.
An example of the forecast errors is presented in Fig. 30 for West Denmark, where the system
operator Eltra is responsible for most of the wind power installed in the area. The wind power
prediction tool in use in year 2001 was dated from year 1997. For Nordpool electricity market
(prediction horizon 13...37 hours ahead) the mean absolute error is 8-9 % of installed
capacity. However, for market operation this results in 38 % of yearly production
mispredicted. For comparison, load is predicted with 1.5…3 % error (mean absolute error, as
% of peak load), which results in about 5 % of yearly energy mispredicted. For prediction 2
49
hours ahead the prediction tools for wind power work significantly better (Fig.30; Holttinen et
al, 2002).
For larger areas the prediction errors decrease. For East and West Denmark, for example, the
errors for day-ahead predictions are to the opposite sides for about a third of time (Holttinen
et al, 2002). For the distance in the direction of most weather systems passing, West-East,
adding East Denmark brings 100 km, or 50 % more to West Denmark.
Eltra. Predictions for 1900 MW wind power, year 2001
duration of prediction
error 13-37 hours ahead
1200
duration of prediction
error 2-3 hours ahead
Prediction error (MW)
900
600
300
0
-300
-600
-900
-1200
1
741
1481 2221 2961 3701 4441 5181 5921 6661 7401 8141
Duration (hours)
Figure 30. Prediction errors for wind power production (state-of-the-art year 1997 for the prediction
tool in use).
50
5 Representative data for large scale wind power
production
To study the impacts of large scale wind power production, the data should be representative
– both in time and space. Depending on what impact we are looking at, we should take an
average year production, or a low or high wind year, to see the extreme situations for system
planning purposes. This means taking production from a representative time period to study.
Depending on what impact we are studying, the wind power production time series should be
representative for the area in question. For example, large scale wind power impacts on the
power system operation, should take the production from large area, with proper smoothing
effect present in the data. This means taking production from representative space.
Checking out the representativeness of time period studied is quite straightforward, when long
term wind power data exists. This is done in chapter 5.1. For checking up the
representativeness in geographical smoothing, no examples were found in the references. In
5.2 some basic parameters from chapter 4 are picked up to form a guideline in this respect.
5.1 Representativeness of the study years
Here we look at the years in question: 1999-2001. Wind power production indices from
national wind power production statistics are presented in Fig. 31 (Laakso, 2003; Carlstedt,
2003; Naturlig Energi, 2003). Wind power production index is a measure of one year’s
production compared with the long term average production. 100 % means that the yearly
production was like the long term average. In Fig. 31 it can be seen that the yearly production
varies between 80...120 %. In Finland, the coastal areas South and West experience a
somewhat different wind resource variations, this is why the production indices are calculated
for 4 sites (Laakso, 2003). The production indices for Finland are here calculated as weighted
average of these indices, using the large scale wind power capacity distribution assumed in
this study (Table 1). For Norway this analysis was not done due to lack of long term data.
However, the Norwegian wind power production seems to experience the same trends as for
the other Nordic countries (Table 2), even if not as strongly.
Year 1999 was less windy than long term average: 86 % in Denmark, 88 % in Finland and 94
% in Sweden (Fig. 31). Year 2000 was close to average (95 % in Denmark, 97 % in Finland,
102 % in Sweden), and year 2001 was clearly less windy than average (80 % in Denmark, 87
% in Finland, 88 % in Sweden).
The production index can be used in determining the long term average wind power
production from only one year of realised production data, by dividing the year’s production
with the year’s index value. For the countries presented here in Fig. 31, we get roughly 25 %
of capacity as the long term average wind power production. The Danish index gives 25.3 %
for both years and Finnish data would give 25.1 % from year 2000 and 25.6 % from year
2001. The Swedish production index is based on wind power production in the South of
Sweden, so using the South Sweden realised production (25.5 % in 2000 and 22.2 in 2001,
51
instead of 26.0 % and 21.9 % respectively) we get 25.0 % from year 2000 and 25.2 % from
year 2001 calculation.
As a total period, 2000-2001 will give a production that is less than average: 88 % of the
average production in Denmark, 92 % in Finland and 95 % in Sweden. Using the years 2000
and 2001 as example years in this study, would give quite good results for Sweden, in that we
are looking at a more windy period (year 2000, 102 % of average production) and a less
windy period (2001, 88 % of average production) compared with long term production. For
Denmark and Finland, year 2000 as a whole was not a high wind year, and as a whole period,
the data in this study will underestimate the wind resource. However, as the data contains also
high wind months, for example the first part of year 2000 (Fig. 15; monthly production
indices in Carlstedt, 2003 and Naturlig Energi, 2003) there are also representative periods of
high wind situations in the data.
Wind power production indices
100 % = average production -87--01
130
DK
FI
SE
120
110
100
90
80
70
60
50
87
88
89
90
91
92
93
94
95
96
97
98
99
00
01
02
year
Figure 31. Yearly wind resource in 1987...2002, according to production statistics of wind energy in
Nordic countries (DK=Denmark, FI=Finland, SE=Sweden).
5.2 Representativeness of the geographical spreading of data
Based on chapter 4 with detailed statistical analyses, it can be estimated, how well the data
represents large scale wind power production. The data used for wind power fluctuations is
critical in the studies for wind power impacts on power system operation. Not to upscale the
fluctuations when upscaling installed wind power in the system, the statistical characteristics
for large-scale production should be looked for in any simulated or meteorological data based
wind power time series (Milborrow, 2001).
As Denmark data is real large scale wind power data of thousands of wind turbines, the
comparisons made in chapter 4 can be used as a basis to estimate how well the data sets
constructed for Norway and Sweden and Finland represent large scale wind power
production.
52
Finland and Norway are considerably larger areas than Denmark, so also the smoothing effect
should be stronger there. For Sweden, there is the possibility of concentrating most of the
wind power capacity south of Stockholm, which means that Sweden should get closer to the
same smoothing effect than in Denmark – probably more if at least some of the capacity was
installed to the Central and Northern part of Sweden.
Summing up the statistical properties for an hourly time series of large scale wind power
production, the following were found in chapter 4:
•
Standard deviation of the hourly production series should be 20–22 % of capacity
for an area like Denmark (300 x 200 km2), if larger area, then less than 20 %
(Finland 18 %, Norway 20 %, Sweden 21 %, Nordic 15 %).
•
Maximum hourly production should be less than 100 %: 85...95 % depending on
how large the area in question is (Denmark 93 %, Finland 91 %, Norway 93 %,
Sweden 98 %, Nordic 85 %).
•
Duration of calms should be non existent or limited (minimum production in
Denmark and Sweden 0 %, Norway and Finland 0.3 and 0.02 %, Nordic 1.3 %).
•
Standard deviation of the hourly variation series should be less than 3 % of capacity
(Denmark 2.9 %, Finland 2.6 %, Norway 3.9 %, Sweden 4.3 % and Nordic 1.9 %).
•
The hourly variations should be in between ± 20 % of capacity, or even less if the
area is larger than the size of Denmark (Denmark -23…20 %, Finland -18…16 %,
Norway -21…27 %, Sweden -32…35 % and Nordic -12…11 %).
The smoothing effect is presented graphically in Figs 10 and 28 where the trends in the
statistical parameters are depicted as a function of the size of the area. Finnish data set is in
line with Denmark data set for reduction of standard deviation, maximum hourly variations
and the standard deviation of variations. Norwegian and Swedish data sets have the statistical
properties above those of Denmark. When looking at the basic statistics for the production
time series, there is not a clear signal that the Norwegian and Swedish data would be
unrepresentative, as taking even few time series from the countries from different locations of
the area gives a basic smoothing effect in the range of production. The analysis on variability,
especially the standard deviation of hourly variations, reveals the caveats of the Swedish and
Norwegian data sets.
The conclusion is that the Finnish data set can be upscaled to represent large scale wind
power production, whereas the Norwegian and Swedish data sets cannot. Combining the 4
data sets to form a Nordic data set probably overestimates the variations somewhat, but a
continuing smoothing effect can be seen (Figs 10 and 28). It has thus been considered
representative for the study of large scale wind power.
There will probably be a slight overestimation of variability for Finnish data when upscaling
the data to large scale wind power production. Even for Denmark, there can be some caveats
as to how well the data represents future wind power production. In the future, there will be
less turbines and sites, but better production from MW scale high turbines, especially for
offshore. When a substantial share of wind energy comes from large offshore wind farms, this
will have an impact on the production, bringing about a less dispersed and thus more variable
53
production, but also higher duration, as there are less calms than on shore (Pryor &
Barthelmie, 2001).
54
6 Wind power production and load
Wind power is a production form that partly resembles electric consumption, the load. It
varies each moment with part of it being unpredictable, causing unexpected variations in the
system. As an example, the wind power production in January, 2000 is presented together
with the load in Fig.32. The wind power production is here upscaled for Finland, to represent
approximately the same wind power penetration level3 (roughly 10 % of gross demand).
Denmark - load and wind power data from January 2000
Wind
7000
Load
6000
MW
5000
4000
3000
2000
1000
0
1
169
337
505
673
Hour
Finland - load and upscaled wind power data (4000 MW, 11 %) January 2000
12000
10000
MW
8000
Wind
Load
6000
4000
2000
0
1
169
337
505
673
Hour
Figure 32. Electricity consumption (load) and wind power production in January 2000. Denmark is
real data (12 % wind power). For Finland data from wind parks is scaled up to wind power
penetration of about 11 % of gross demand.
The y-axis scale in the graphs is not the same, but the scale is relatively the same in the sense
that it shows the load in these countries when the 10 % wind production curve is on the same
3
The share of produced wind power in the power system is here denoted as wind power penetration. The share is
presented here as % of energy, yearly gross demand. Penetration as % of installed capacity is also used in some
studies, which is a considerably larger figure than expressing it as % of energy.
55
level. It can be seen that the load variations in Denmark are relatively larger than in Finland,
with an energy intensive industry.
In this chapter the basic patterns of electrical load, together with wind power production, are
presented. The main focus is on the hourly variations and on peak load situations.
6.1 Basic statistics of the hourly load time series
Time series of load in Nordic countries, featuring also duration curves, are presented in Fig.33
for year 2001, and for year 2000 in Appendix 3. Electric load is characterised by a daily
pattern, higher on weekdays than weekends (Seppälä, 1996). In addition to daily cycles,
temperature effects can be seen in the graphs: the load is generally lower during summer, and
different weeks in wintertime show dependence on temperature. For Finland, the holidays in
midsummer and Christmas mean more pronounced dips in consumption than for the other
countries. Even as the y axis scale is different in the graphs, as the scale is relative to the peak
load, it can be seen that the load varies relatively more in Denmark compared with other 3
countries with energy intensive industry. Also widely used electric heating in Sweden and
Norway and to a lesser extent in Finland can explain the differences. There are no major
differences between the years, except for Finland where a strike in the energy intensive forest
industry dropped the load by about 3000 MW for one week in April 2000 (Appendix 3).
Basic statistics of the load time series is presented in Table 10 for the years 2000 and 2001. In
both Sweden and Norway the consumption is larger than Finland and Denmark together.
Denmark has by far the lowest consumption, only about 10 % of the total Nordic demand. It
can be seen, that electric consumption has been rising by 3 % between the years (only 1 % in
Denmark, and 4 % in Finland and Sweden). The peak load has risen even more, but this is
probably due to year 2000 not having extreme cold weather periods. The peak load is about 3
times larger than the minimum load. Some smoothing can be seen in the total Nordic load
time series, the peak is lower and minimum load higher than the sum of the countries, as the
peaks do not coincide. The Finnish load series is considerably less variable than for the other
countries, as can be seen from the standard deviation relative to mean value.
Table 10. Key figures for electric load during example years 2000 and 2001.
Denmark
Finland
Norway
Sweden
Nordic
2000
2001
2000
2001
2000
2001
2000
2001
2000
2001
sum (TWh)
34.82
35.18
76.19
79.11
120.05
123.32
140.83
147.26
371.90
384.87
Peak
6284
6229
11829
12579
20420
23054
25381
26323
62265
67854
min
1964
2062
3571
3537
6832
8289
9042
9157
24586
23838
3.2
3.0
3.3
3.6
3.0
2.8
2.8
2.9
2.5
2.8
3964
4016
8674
9031
13667
14078
16033
16810
42338
43935
937
927
1200
1409
2724
3327
3254
3791
7681
9076
23.1 % 13.8 %
15.6 %
19.9 %
22.6 % 18.1 %
20.7 %
peak/min
average
stdev
stdev/average
23.6 %
56
23.6 % 20.3 %
Nordic 2001
80000
hourly load (MW)
70000
60000
50000
40000
30000
20000
10000
0
1
721
1441
2161 2881
3601
4321 5041
5761 6481
7201
7921 8641
5761
7201
7921
hour
Finland 2001
14000
hourly load (MW)
12000
10000
8000
6000
4000
2000
0
1
721
1441
2161
2881
3601
4321
5041
6481
8641
hour
Sweden 2001
30000
hourly load (MW)
25000
20000
15000
10000
5000
0
1
721
1441
2161
2881
3601
4321
5041
5761
6481
7201
7921
8641
hour
Denmark 2001
7000
hourly load (MW)
6000
5000
4000
3000
2000
1000
0
1
721
1441
2161
2881
3601
4321
5041
5761
6481
7201
7921
8641
hour
Figure 33. Hourly load of the Nordic countries, chronologically and as duration curves. The y-scale is
different for all the graphs.
57
The total electrical consumption in the hourly time series (Table 10) is not exactly measured.
This is why the electricity statistics show slightly different values (Nordel, 2003): for year
2001: 35.43 for Denmark, 81.52 for Finland, 120.36 for Norway, 148.91 for Sweden, a total
of 386.22 TWh.
6.2 Correlation of load and wind power
Correlation with electrical load is important, when considering the power system effects of
wind power. When there is a diurnal pattern in wind power production, coinciding with the
load, like wind power production increasing in the morning and decreasing in the evening,
this has a beneficial effect. In Table 11, the correlation between the electric load and wind
power production is shown.
Table 11. Correlation between the electric load and wind power production in Nordic countries, years
2000 and 2001.
SE load
NO load
FI load
WestDK
load
EastDK
load
DK load
Nordic load
SE load
1.00
NO load
0.96
1.00
FI load
0.88
0.83
1.00
WestDK load
0.75
0.64
0.67
1.00
EastDK load
0.86
0.77
0.75
0.95
1.00
DK load
0.80
0.70
0.71
0.99
0.98
1.00
DK wind
0.20
0.20
0.22
0.20
0.22
0.21
0.21
FI wind
0.13
0.15
0.16
0.08
0.11
0.09
0.14
NO wind
0.34
0.37
0.35
0.15
0.23
0.18
0.35
SE wind
0.24
0.25
0.25
0.20
0.22
0.21
0.25
Nordic wind
0.30
0.32
0.32
0.21
0.26
0.23
0.31
The upper part of the table shows the correlation between the hourly loads in the countries.
They are strongly correlated inside Denmark and between Norway and Sweden and less than
strongly correlated between West Denmark and Norway/Finland. Below, the correlation
between the load and wind power production is shown. For the whole data it seems that the
correlation is slightly positive, but when looking at the 3 winter months only (Table 12), the
correlation is zero or even slightly negative. Negative correlation means that there is a slight
tendency for the wind power to decrease when load is increasing, and vice versa. For the load
time series, there is a slight reduction in the correlation coefficients wintertime, compared
with the values calculated for the whole data.
58
Table 12. Correlation between the electric load and wind power production in Nordic countries
during winter months January, February and December, 2000 and 2001.
SE load
NO load
FI load
WestDK
load
EastDK
load
DK load
Nordic load
SE load
1.00
NO load
0.92
1.00
FI load
0.78
0.72
1.00
WestDK load
0.82
0.73
0.64
1.00
EastDK load
0.85
0.76
0.65
0.97
1.00
DK load
0.84
0.75
0.65
0.99
0.99
1.00
DK wind
-0.11
-0.13
0.01
0.09
0.04
0.07
-0.07
FI wind
-0.18
-0.17
-0.07
0.01
-0.01
0.00
-0.14
NO wind
-0.13
-0.11
-0.04
-0.01
-0.03
-0.02
-0.10
SE wind
-0.10
-0.10
0.02
0.07
0.03
0.06
-0.06
Nordic wind
-0.16
-0.15
-0.02
0.06
0.01
0.04
-0.11
6.3 Temperature dependence
In Northern Europe, the electricity demand is strongly correlated with outdoor temperature.
This is why the correlation between wind power production and temperature is also relevant
for the adequacy of power production, when determining the capacity value of wind power.
There is a clear negative correlation between the load and temperature (about –0.7), which
means that when temperature drops the load increases. The only exception is Denmark, where
the correlation is very weak. There is also a slight negative correlation between wind power
and temperature, but for Denmark and Finland it is close to zero (Table 13).
In Figs 34-36, the temperature dependence is depicted for Finland and Denmark, as well as
for the Nordic wind power as the function of Finnish temperature. The average wind power
production at low temperatures of below –15 oC is somewhat lower than average in Finland,
and these are the incidents of highest load (Fig. 34). The average wind power production in
Denmark as well as the total Nordic wind power does not experience this kind of reduction
(Fig. 35).
59
Table 13. Table temperature correlation of load and wind power production in the Nordic countries,
hourly data from 2000-2001
FI
DK East
NO
temperature temperature temperature
FI temperature
1,00
DK East temperature
0,88
1,00
NO temperature
0,88
0,82
1,00
FI load
-0,63
-0,56
-0,57
SE load
-0,72
-0,68
-0,69
NO load
-0,79
-0,76
-0,79
DK load
-0,25
-0,21
-0,27
Nordic load
-0,72
-0,67
-0,69
FI wind
-0,09
-0,11
-0,11
SE wind
-0,18
-0,14
-0,20
NO wind
-0,32
-0,32
-0,28
DK wind
-0,13
-0,07
-0,15
Nordic wind
-0,32
-0,32
-0,28
max wind power
min wind power
average wind power
max load
100 %
% of capacity / peak load
90 %
80 %
70 %
60 %
50 %
40 %
30 %
20 %
10 %
0%
-27 -24 -21 -18 -15 -12 -9 -6 -3
0
3
6
9 12 15 18 21 24 27 30
Temperature at 2 m, West coast (C)
Figure 34. Temperature dependence of wind power production
and load in a cold climate, example
fromo Finland. There were 48 hours (0.2 % of time) below –23 oC and 489 hours (1.9 % of time) below
–14 C during the study years 1999-2001.
60
75000
67500
60000
52500
45000
37500
30000
22500
15000
7500
0
-25
-20
-15
-10
-5
0
5
DK wind ave
Nordic max wind
10
15
20
1
0,9
0,8
0,7
0,6
0,5
0,4
0,3
0,2
0,1
0
wind power production (% cap)
Load (MW)
Nordic load max
Nordic wind ave
25
Temperature west coast Finland (C)
DK load max
DK wind ave
Load (MW)
7000
count
Nordic wind ave
0,7
6000
0,6
5000
0,5
4000
0,4
3000
0,3
2000
0,2
1000
0,1
0
wind power production (% cap)
Figure 35.Wind power production and load in Nordic countries as a function of temperatures in
Finland. Years 2000-2001.
0
-13
-8
-3
2
7
12
17
22
27
Temperature Denmark East (C)
Figure 36. Temperature dependence of wind power production and load in Denmark. Years 20002001.
6.4 Instant penetration level of wind power
When wind power penetration is considerable, that is, wind energy produces more than 5 or
10 % of the yearly gross demand, the wind power production peaks can be clearly seen by the
system. To illustrate this, instant penetration has been looked at. This is the wind energy share
of a single hour’s load. This is relevant for high penetrations of wind power, when some wind
energy has to be discarded in order to keep the power system stable. Example for instant
penetration data is illustrated for Denmark year 2000 data, with wind power upscaled to 4000
61
MW, 24 % of yearly gross demand (Fig. 37). In Table 14, periods of high instant penetration
for years 2000 and 2001 are shown, upscaling the wind power production in Finland and
Denmark to different levels.
Table 14. Penetration of wind power – maximum hourly penetration levels are more than 5 times
higher than the average yearly value for Denmark. When the yearly penetration exceeds 20 % of gross
demand, there are hours when wind power production exceeds the load (penetration > 100 %).
year
2000 DK, MW
installed
capacity
MW
% of
maximum >50 %,
>70 %,
>90 %,
>100 %,
yearly penetration number of number of number of number of
demand
hours
hours
hours
hours
1740-2200
12.2 %
65 %
28
0
0
0
2000 DK, % cap
2000
12.2 %
58 %
40
0
0
0
2000 DK, % cap
3000
18.3 %
87 %
458
77
0
0
2000 DK, % cap
4000
24.3 %
116 %
1321
380
106
40
2001 DK, % cap
2000
10.1 %
58 %
24
0
0
0
2001 DK, % cap
3000
15.2 %
87 %
361
48
0
0
2001 DK, % cap
4000
20.2 %
116 %
891
277
66
24
2001 DK, % cap
5000
25.3 %
146 %
1422
691
236
138
2000 Finland
4000
11.2 %
53 %
3
0
0
0
2000 Finland
6000
16.9 %
79 %
91
4
0
0
2000 Finland
8000
22.5 %
105 %
654
58
6
3
2001 Finland
4000
9.8 %
39 %
0
0
0
0
2001 Finland
6000
14.7 %
59 %
49
0
0
0
2001 Finland
8000
19.6 %
79 %
467
18
0
0
2001 Norway
6000
13.4 %
50 %
0
0
0
0
Instant penetration from Danish real data, in MW, was 65 % in 2000 (wind 12.2 % of gross
demand). The few peaks over 60 % occurred in November Sunday night, when load dropped
to below 3000 MW and wind was about 1800 MW. As the capacity rose for about 300 MW
during the year, and the instant penetration was reached at the end of the year, this 65 % value
represents actually a situation where the yearly penetration level is more than 12 % of yearly
penetration level. When looking at the Danish data as % of capacity, putting 2000 MW
capacity which gives the same yearly penetration of just over 12 %, the maximum instant
penetration is less than 60 %.
In West Denmark, already some hours of 100 % instant penetration of wind power have been
recorded. The integration problem in Denmark involves distributed, small combined heat and
power plants. This is because during cold windy winter periods the production from these
weather dependent production forms sometimes exceeds the load and available transmission
to neighbouring countries, when a part of the centralised thermal plants have to be in
operation to provide balancing and stability for the system (Eriksen et al, 2002).
62
DK wind 4000 MW
7000
DK load
6000
MW
5000
4000
3000
2000
1000
0
1
49
97 145 193 241 289 337 385 433 481 529 577 625 673
hour
Figure 37.Example of high instant penetration level of wind power. Denmark, year 2000 data, wind
power upscaled to 4000 MW corresponding to 24 % of yearly gross demand. During
October/Novermber, wind power production equals load during some hours in the night time.
6.5 Wind power during peak load
When studying the impact of wind power on the adequacy of power system, wind power
production during the peak load periods is of crucial importance. Wind power production
during the 10, 50 and 100 highest peak load hours during the example years is shown in Table
15.
Table 15. Wind power production, as % of installed capacity, during highest peak load hours.
Denmark 2000
Denmark 2001
Finland 1999
Finland 2000
Finland 2001
Sweden 1999
Sweden 2000
Sweden 2001
Norway 1999
Norway 2000
Norway 2001
Nordic 2000
Nordic 2001
The whole year
Average (min-max)
24.1 % (0.0-92.7 %)
20.3 % (0.0-90.1 %)
21.6 % (0.0-86.3 %)
24.4 % (0.1-91.1 %)
22.2 % (0.0-86.1 %)
24.7 % (0.0-100 %)
26.0 % (0.0-98.5 %)
21.9 % (0.0-95.6 %)
29.3 % (0.3-95.5 %)
33.8 % (0.3-93.1 %)
31.4 % (0.3-92.6 %)
27.1 % (1.3-84.9 %)
23.4 % (1.6-90.2 %)
During 10 peaks
During 50 peaks
During 100 peaks
Average (min-max) Average (min-max) Average (min-max)
24.0 % (0.7-69.6 %) 30.5 % (0.5-86.5 %) 31.4 % (0.4-86.4 %)
36.6 % (0.2-74.4 %) 30.4 % (0.1-87.1 %) 27.8 % (0.1-87.1 %)
6.9 % (4.7-10.2 %)
7.1 % (2.9-36.9 %)
8.5 % (2.2-45.9 %)
36.1 % (4.4-72.4 %) 31.7 % (3.4-72.2 %) 28.7 % (3.4-75.2 %)
18.5 % (3.1-37.6 %) 19.3 % (3.0-38.2 %) 17.5 % (3.0-38.2 %)
22.8 % (15.8-29.1 %) 20.3 % (1.9-62.8 %) 20.4 % (0.8-66.2 %)
15.8 % ( 4.5-49.8 %) 16.9 % (0.5-57.6 %) 17.6 % (0.1-77.6 %)
60.7 % (52.7-66.6 %) 35.5 % (1.4-66.6 %) 29.8 % (1.4-73.0 %)
52.1 % (35.4-81.7 %) 48.5 % (14.1-81.9 %) 42.3 % (9.5-81.9 %)
35.6 % ( 8.8-74.5 %) 34.8 % (8.8-74.5 %) 35.2 % (8.8-79.1 %)
60.8 % (38.5-84.4 %) 54.5 % (25.5-84.4 %) 46.5 % (15.3-84.4 %)
16.4 % ( 4.1-42.1 %) 21.1 % (4.1-56.8 %) 24.4 % (4.1-71.3 %)
45.5 % (41.8-50.3 %) 35.3 % (7.5-57.1 %) 28.8 % (7.1-57.1 %)
63
As wind power production from one area can be zero during wintertime, it is often assumed
that wind power does not contribute to the adequacy of power production system. The socalled capacity credit of wind power is neglected. Capacity credit means the capacity that
wind power can be counted on with a similar probability than conventional power units.
Determining the capacity credit for wind power involves estimations of the production over a
long time period, at least 10 years, as well as simulations of the system loss-of-load
probability. It is therefore out of the scope of this study. However, with the hourly data from
Nordic countries, it is interesting to look at the wind power production during peak load
hours, based on this realised wind power production.
Years 1999 and 2001were low wind years and year 2000 was a near average wind year
(chapter 5.1). However, this reflects the production during the peak load hours only for
Finland. For the whole Nordic area, the production during peak load was higher than the
yearly average power in 2001 and less than average in 2000.
In 1999, the highest peak occurred on 29th January, both in Finland and Sweden. The 10
highest peaks were all in two days, 28th and 29th. In Sweden (data only for South Sweden), the
wind power production was close to average, but in Finland wind power production was low,
less than 10 % of capacity. In Norway, the highest peak was on 15th December, all the highest
peaks also within two days 15th and 16th. Wind power production (data only for Middle and
North Norway) was close to average on the highest peak and higher than average during the
other 9 highest peak load hours.
In 2000, the eight highest peak load hours in Sweden were on 24th January, when the wind
was less than 10 % of capacity. The peaks number 9 and 10 were on 21st January, when the
wind was higher than average. For Finland the highest peaks occurred on three days (21st, 24th
and 25th January). For Norway and Denmark there were several days in January and
December for the highest load hours, so the wind resource had also a wider range.
In 2001, the 10 highest peaks were all on 5th February in Sweden, when the wind was strong.
For Norway, the peaks were during 5th and 6th, also with strong winds. For Finland, the peaks
were during 3rd, 5th and 6th and the winds were part of the time low (5th morning hours) and
part of the time near average production (5th evening hours and 3rd and 6th).
The average wind power production during the peak load situations (100 highest peaks) is
close to average production over the year, with the exceptions of Finland 1999 and Sweden
2000 with lower production and Norway 1999 and 2001 for higher production. All in all, the
table 15 gives some evidence that wind power can be counted on, with some probability
greater than 0 being available during peak loads. This is also the result of several studies of
wind power capacity credit (Milligan, 2000; Giebel, 2001). In Norway, previous study of
wind speed data gave the result that the probability of delivering wind power during peak load
was lower than on average during wintertime, but higher than on average during the whole
year. Distributing the production increased the probability of wind power production during
the peaks (Alm & Tallhaug, 1993).
64
6.6 Hourly variations of load
The hourly variations of load follow a clear diurnal cycle (Fig. 38). The typical range of daily
cycle can be estimated from the Fig.32. It is 16 000 MW for the total Nordic load, nearly 2500
MW for Denmark, 2000 MW for Finland, for Norway 2000 MW in summer and 4000 MW in
winter, and for Sweden 4000 MW in summer and 6000 MW in winter.
Load
7000
Hourly variation
6000
5000
MW
4000
3000
2000
1000
0
-1000
1
25
49
73
97
121
145
169
Hour
Figure 38. Hourly time series of the load and its variations, Denmark first week of January, 2000.
Example of the hourly variations of load is presented in Fig.39 for Finland year 2000, for
other countries and year 2001 in Appendix 4.
hourly load variation (MW)
1500
1000
500
0
-500
-1000
-1500
1
741
1481
2221
2961
3701
4441
5181
5921
6661
7401
8141
Figure 39. Hourly load variations, example Finland 2000, chronological time series and duration
curve.
Basic statistics of hourly variations are shown in Table 16. Although the load in Finland is
almost twice the one in Denmark, the variations of load are nearly the same in MW.
65
Table 16. Hourly variations of load in the Nordic countries.
Nordic
Denmark
Finland
2000
2001
2000
2001
2000
2001
max down-variation (% of peak)
-7.8 %
-5.4 %
-13.5 %
-13.4 %
-8.3 %
-7.2 %
max up-variation (% of peak)
10.8 %
9.0 %
18.1 %
18.3 %
9.7 %
8.2 %
-4866
-3642
-849
-838
-985
-900
max up-variation (MW)
6698
6081
1140
1141
1144
1035
standard deviation (MW)
1446
1430
278
271
263
269
max down-variation (MW)
The range of hourly variations is inside ± 10 % of peak load for the total Nordic load and for
Finland, for Denmark –13...+18 % of peak load. The longer variations are more than double.
The maximum 12-hourly variations for Denmark are about ±2300 MW (up and down) in
summer and ±2600 MW (up and down) in winter. The maximum 4-hourly upward variations
are about the same, but the downward variations are –1400 MW down in summer and –1600
MW down in winter. For Finland, there is more difference between the years 2000 and 2001
for the maximum variations. The maximum 4-hourly variations are 2200...2400 MW up. The
maximum downward variations are –1700 MW in summer and –2000 MW in winter. The
maximum 12-hourly variations are 2600...2800 up and –2200... –2800 MW down. These are a
bit larger values than the estimated typical diurnal variations, from the graphs in Fig. 33.
66
7 Increase in net load variations by wind power
The additional requirements and costs of balancing the system in the operational time-scale
(from several minutes to several hours) are primarily driven by fluctuations in wind
generation output. Part of the fluctuations are predictable 2...40 hours ahead. The varying
production pattern of wind power is changing the scheduling and unit commitment of the
other production plants and use of transmission between regions – either losses or benefits are
introduced to the system, compared with the situation without wind. Part of the fluctuations
remain unpredicted, or mispredicted. This is what has to be handled by regulation market and
balancing services (primary and secondary reserves).
To estimate the impact of wind power on power system operational reserves, it has to be
studied on a control area basis. Every change in wind output does not need to be matched onefor-one by a change in another generating unit moving in the opposite direction. It is the total
system aggregation that has to be balanced.
When small amounts of wind power are added to the energy system, there will be minor
effects on the system behaviour. The system is dimensioned to cover the varying load,
electricity consumption, at every instant. Wind power production can be seen by the system as
negative consumption – when wind power is produced, the varying load that the system will
see, the net load, will be sometimes slightly lower and sometimes the same as without wind.
When the amount of wind is small, the effect to the net load cannot be distinguished from the
difference of real load versus load prediction in normal operation of the system.
The need for regulating and load following capacity in the system increases when wind
production causes larger variations to the system than the variations of existing load. The need
for more flexibility in order to meet larger fluctuations in the system depends on how much
wind there is in the system – what portion of consumption is covered by wind production.
Also systems are different: the amount of load variations as well as the flexibility in the
system differ from country to country.
In Fig.40, the same time series as in Fig.32 are shown for January, 2000, but the wind power
production is subtracted from the load to show the effect of wind on the variations that the
system will see. As the load in Finland varies considerably less than in Denmark, a 10 %
penetration of wind would result in larger changes in the system in Finland than in Denmark.
As the scale in Fig.40 is one month, 740 hours, mainly the longer term variations (12-48 h),
and the changes in those, can be seen. In longer time scales there is time for the system to
react to these changes – it is the time scale of electricity markets. It is clear from Fig.40, that
to accommodate larger shares of wind power, good prediction models for wind power
production are needed.
The short term variations were studied by hourly time series. Variations within an hour are
less than the hourly variations (chapter 4.7). Therefore, hourly variations can be used as an
estimate for short term variations. Because the quality of data from Norway and Sweden were
judged insufficient for the hourly variation studies (chapter 5.2), the study is here made for
Denmark, Finland and the total Nordic data series.
67
Denmark - load and wind power data from January 2000
7000
6000
MW
5000
4000
3000
2000
Load
1000
Load - Wind
0
1
169
337
505
673
Hour
Finland - load and upscaled wind power data (4000 MW, 11 %), January 2000
12000
10000
MW
8000
6000
4000
Load - Wind
2000
Load
0
1
169
337
505
673
Hour
Figure 40. Electricity consumption (load) and the net load (wind production subtracted from load) for
2000 MW wind power in Denmark and 4000 MW wind power in Finland.
The net load hourly variations are calculated like the hourly variations (4), but now for the net
load time series, where wind power production is subtracted from load:
∆NLi = NLi − NLi −1 = ( Li − Pi ) − ( Li −1 − Pi −1 ) = ∆Li − ∆Pi
,
(5)
where NL denotes the net load (MW), L the load (MW) and P the wind power production and
i is the hour (from 2...8760 in 2001 and 2...8784 in 2000).
In Fig.41, the amount of hourly variations that the system sees is depicted, without wind (the
hourly variations of the load) and with wind (the hourly variations of net load). The difference
in the maximum value indicates the amount that the operating reserve capacity has to be
increased. The difference in the duration curves indicates the amount that the existing reserve
capacity is operating more, when wind power is added. Same capacity can in principle be
used for both up and down regulation, and the variations as well as the increase should
basically be symmetrical. Either up or down variations can determine the need of increase in
68
the reserves. In many systems, it is the up regulation that is more critical to handle by the
system.
The increase in hourly variations due to wind power is in the following chapters estimated in
3 ways. The increase in hourly variations can be taken as an estimate for increase in
requirement for load following or secondary reserve in the system. The results are
summarised in Table 17 at the end of 7.3.
1500
FI load variations
FI net load variations
variation, MW
1000
500
0
-500
-1000
-1500
1
741
1481
2221
2961
3701
4441
5181
5921
6661
7401
8141
hour
Figure 41. Duration curve of load variations (without wind power) and net load variations (load-wind
power) example Finland, year 2000, 6000 MW wind (17 % of gross demand).
7.1 Wind power increasing the largest hourly variation in the
system
Variations in net load (load minus wind production) compared to variations in load give an
estimate for the needs of the system to react to large scale wind power on short term. Wind
has an effect on the total amount of load following reserve capacity, if the maximum of net
load variations is larger than the maximum of load variations. The largest difference in hourly
variations was looked for. This is the increase of variations that the system will see.
The results for 2 years and two countries, Finland and Denmark, are presented in Fig. 42, for
both the maximum upward variation (increase in downregulation) and maximum downward
variation (increase in upregulation). Upscaling the wind power production and looking for the
increase of maximum hourly variation in the net load time series, the curves are sometimes
increasing linearly and sometimes piecewise linearly depending on what was the wind power
variation in relation to the critical few hours of largest load variations. It can be seen from
Fig.42, that this kind of analysis is very sensitive to the hourly data in question and can give
69
increase in maximum variation
(% cap)
very different results for different years. The increase in variations can be 0...4 % at 5 %
penetration, 0...5.5 % at 10 % penetration and 2...7 % at 15 % penetration.
10 %
FI 2000 increase in downreg
8%
FI 2000 increase in upreg
FI 2001 increase in downreg
6%
FI 2001 increase upreg
4%
DK 2000 increase in downreg
DK 2000 increase in upreg
2%
DK 2001 increase in downreg
0%
-2 %
0%
DK 2001 increase in upreg
5%
10 %
15 %
20 %
25 %
30 %
wind power penetration (% of gross demand)
Figure 42. Maximum hourly variation of net load time series compared to load time series gives the
increase in variations seen by the power system. Example from upscaling wind power production data
for Denmark and Finland.
7.2 Wind power increasing the hourly variations in the system
Looking at one single maximum hourly variation per year when determining the increase in
the variations due to wind can overestimate the effect, especially if there is any doubt in the
reliability of the data. For example, the hourly load data is based on measurements of most of
the production and transmission, and part of it is estimated. This is why the largest hourly
variations can be due to erroneous data.
Reserves in the power system are often determined using a probabilistic approach – to prepare
for variations that are within certain limits of probability, for example covered with 99.99 %
probability.
For a normally distributed probability distribution, it is known how much data is within the
range of ±σ of the mean value. Taking a range of ±3σ will cover 99 %, and ±4σ will cover
99,99 % of all variations. For hourly variations, the mean value is 0.
From Fig.28 and table 16 in chapters 4 and 6 for basic statistics of wind power and load time
series respectively, the standard deviation of the hourly variations can be seen. As the
variations of load and wind power production can be assumed uncorrelated4, the standard
deviation of net load time series can be determined by a simple square root sum of the
standard deviations of load and wind power time series:
4
The use of formula (6) was checked for this data and it produced accurate results for the standard deviation of
the net load
70
σ NL 2 = σ L 2 + σ W 2
(6)
Finally, the increase in the variations can be formulated as the increase in 4σ variations (Fig.
43):
I = 4(σ NL − σ L )
(7)
Calculating this way, I am assuming that wind power only contributes to the load following
requirement by the increase due to its addition to the system, that is, wind power gets the
benefit of the existing power system. In USA, different allocation methods have been
elaborated (Kirby & Hirst, 2000), where the benefit of joining two varying elements is
divided by the two, in this case the system would benefit a part of the addition of wind power.
This would demand more from wind power than the simple increase in variations calculated
here by formula 7. Both methods are numerically correct, it is a question of fairness or design
of regulation payments. In the Nordic countries, different loads and production units do not
pay different tariffs for the regulation burden they pose to the system. Until the reserve
requirements are allocated to loads and production units it is well justified to calculate only
the simple addition to reserve requirements for wind power.
2500
FI load var
FI net load var
Frequency (count)
2000
1500
4σL =1063 MW
1000
4σNL =1139 MW
500
152 MW
52 MW
0
1500
1300
1100
900
700
500
300
100
-100
-300
-500
-700
-900
-1100
-1300
-1500
Hourly variations (MW)
Figure 43.An example of estimating the increase in hourly variations seen by the system. If only
maximum variation is looked at, the increase is determined at the tails of the distribution (52 MW
increase in up- variation and 152 MW increase in down-variation). Looking at the standard deviation
of the distributions, there is a difference of 76 MW in the 4σ coverage of the variations.
71
increase in hourly variations
(4sigma) (% cap)
5%
4%
FI
3%
DK
Nordic
2%
1%
0%
0%
5%
10 %
15 %
20 %
25 %
30 %
wind power penetration (% of gross demand)
Figure 44. Increase in hourly load following requirement for wind power. Increase is relative to
installed wind power capacity.
The probabilistic approach gives lower requirements than only looking at the maximum
changes. The increase in variations is 0.5–1 % of installed wind power capacity at 5 %
penetration (of gross demand), 1–2 % at 10 % penetration and 1.7–2.7 % at 15 % penetration.
In other words, 2000 MW in Denmark increases the variations by 20 MW and the same
penetration level for Finland, 4000 MW increases the variations by 80 MW. The reason why
the effect of wind power on variations is smaller in Denmark than in Finland is mainly based
on the relatively larger load variations in Denmark, absorbing wind variations. Finnish wind
power data used here will overestimate the hourly variations of wind power to some extent,
because real production from thousands of units giving 10 % of yearly electricity would see
less variations than the data upscaled from 57 turbines data used here.
The same analysis was also made to the combined time series representing the Nordic wind
power production. If the Nordic market area was working without bottlenecks of
transmission, also the short term variations of wind power could be absorbed by the system. If
the total wind power production was distributed evenly to the 4 countries, this would result in
increased hourly variations in the system than the load variations today, of less than 1 % of
installed capacity at 10 % wind penetration (of gross demand). In other words, 19 000 MW of
wind power in the Nordic countries would increase the hourly load following requirements by
about 160 MW. Also this is a conservative estimate, as half of the data, for Sweden and
Norway, exaggerate the hourly variations.
7.3 Wind power increasing the unexpected hourly variations of
load
The analysis in previous chapter 7.3 assumes that the hourly variations of both load and wind
power production are unexpected. However, as the load with its clear diurnal pattern is easier
to forecast than wind power production, this should be taken into account when analysing the
increase in operating reserve requirement due to wind power (Milligan, 2003).
72
For wind power, the production an hour ahead can be reasonably well forecasted by
persistence, that is, taking the production level at hour i-1 for the predicted value at hour i
(that is, using the hourly variation as used in previous chapters 7.2 and 7.1). The short term
prediction tools can to some extent improve on this, taking into account the forecasted trend
of wind speeds in the area, as well as time series techniques that have proven to work quite
well for some hours ahead. The persistence is therefore a conservative estimate for the wind
power production an hour ahead.
The load prediction has been studied for decades, it is well known and the predictions are
quite accurate (within 1-2 %). There is a diurnal pattern and dependence of temperature in the
demand for electricity. A case study for Finland year 2001 load data was carried out to
estimate load forecasts. A model at VTT was used, based on calendar days of loads (from
year 2000 data) and temperature (Koreneff et al, 1999; Koreneff & Kekkonen, 2000). The
mean absolute error, hour ahead, was 0.7 %. This is probably lower than what is experienced
in different system areas in an average (Milligan, 2003). The forecast error for the load was
then compared to wind power variations. The standard deviation of forecast error was 123
MW (1 % of peak load), in comparison of 267 MW for the load hourly variations, so this
method assumes that about half of the variability in load can be predicted.
Now making the same analysis as in 7.2, but using load forecast error instead of the hourly
variation of load, we get the results in Table 17 for different wind power prediction error
levels.
Table 17. Summary of results for the increase in hourly variations by wind power in Finland. For
maximum hourly variation: if positive, the value is increasing from last hour to current hour.
Wind power MW
Wind power penetration % of gross
demand
maximum hourly variation of wind MW
maximum hourly variation of load MW
maximum hourly variation of net load MW
Increase in maximum hourly variation MW
stdev wind power hourly variations MW
stdev load hourly variations MW
stdev net load hourly variations MW
Increase in variations, 4σ, MW
stdev load forecast error MW
Increase in forecast error variations, load
forecast only, 4σ, MW
stdev wind forecast error MW
Increase in forecast error variations, 4σ,
MW
2000
4000
6000
4,9 %
280 / -310
1144 / -985
1138 / -1061
-6 / 76
52
266
271
20
123
9,7 %
560 / -620
1144 / -985
1191 / -1137
47 / 152
103
266
285
77
123
14,6 %
840 / -930
1144 / -985
1385 / -1214
241 / 229
155
266
308
167
123
41
41
150
82
298
124
27
100
206
The results in Table 17 show that the results in previous chapter 7.2, based on the simple
hourly variations from load and wind power time series, should be increased with 50-100 %
depending on the level of wind power forecast (no forecast to hour ahead ... forecast being 20
% better than not using any). This means that when producing 10 % of yearly electricity
consumption with wind power, the increase in hourly load following requirement would be
73
1.5–4 % of the installed wind power, instead of 1-2 % as the result of previous chapter 7.2.
More specifically, for Denmark 2000 MW of wind power would increase the load following
requirement by 30-40 MW, for Finland the 4000 MW with 100–150 MW and for the Nordic
countries the 19 000 MW with 240–320 MW.
74
8 Summary and conclusions
The usual drawbacks of wind power from the power system point of view are that wind power
production is variable, difficult to predict and cannot be counted for. However, problems of
integrating intermittent sources are reduced when they are connected to large power systems,
which can take advantage of natural diversity in variable sources. Large geographical
spreading of wind power will reduce variability, increase predictability and decrease the
occasions with near zero or peak output.
High wind power penetration will increase the flexibility needed in the system. The
magnitude of the power system impacts of wind power depend on how large a share is
produced by wind power, as well as on the power system in question. When we are studying
the incremental effects that varying wind power production imposes on the power system, it is
important to study the system as a whole: only the net imbalances have to be balanced by the
system.
In this study, the focus is on the hourly time scale impacts on the power system, based on real
wind power production and synchronous hourly load data. Example years of 2000 and 2001
were studied. Year 2000 was close to average wind year (95 % of long term average
production in Denmark, 97 % in Finland and 102 % in Sweden). Year 2001 was clearly less
windy than average (80 % of long term average in Denmark, 87 % in Finland and 88 % in
Sweden).
Average production in the Nordic countries is highest in Norway (31–34 % of installed
capacity), and about 22–24 % of capacity for the other countries during the example years.
The seasonal variation was clearly present in the data sets, more production in winter than in
summer. Wind power production in Denmark and Sweden experience a more pronounced
diurnal variation, whereas the sites in the northern part of Finland, Sweden and Norway do
not experience any detectable diurnal variation.
From the combined production in the Nordic countries, it can be seen that as wind power
production comes from geographically distributed wind farms, the total production never
reaches the total installed capacity and it is hardly ever totally calm. Production above 50 %
of rated capacity is rare in summer and production above 75 % is rare in winter. The lowest
hourly production was 1.3 % of capacity. The production was below 5 % of capacity about 2
% of time. There was not a significant difference between the calm periods in years 2000 and
2001. For the peak production, defined as above 75 % of capacity, there were more peaks
during year 2000 than during year 2001.
Correlation for hourly wind power production is strong (more than 0.7) for distances of less
than 100 km and becomes weak (below 0.5) with distances above 200-500 km. The large
scale wind power production of the countries is correlated between Denmark and Sweden,
and weakly correlated between the other countries. No correlation between the hourly
variations of wind power production in the countries was seen in the data.
75
The hourly variations of large scale wind power production are about 90 % of time between ±
5 % of capacity and 99 % of time between ± 10 % of capacity. For the total Nordic time series
the hourly variations are about 98 % of time between ± 5 % of capacity.
The representativity of the constructed wind power data sets for Finland, Norway and Sweden
was estimated based on the statistical properties of existing large scale wind power production
data from Denmark. An hourly time series of large scale wind power production should have
standard deviation of the production series less than 20 % of capacity, maximum hourly
production less than 100 % (85...95 % depending on how large the area in question is),
duration of calms limited or non existent, standard deviation of the hourly variation series less
than 3 % of capacity and the hourly variations in between ±20 % of capacity, or even less if
the area is larger than the size of Denmark (300 x 200 km2). According to these criteria, the
data set for Finland is quite representative for large scale wind power production, but the data
sets for Norway and Sweden are not. This is mainly discovered by the hourly variations of the
production time series which is not as smooth as a large scale wind power production from
thousands of turbines would be. Combining the 4 data sets to form a Nordic data set probably
overestimates the variations some, but a continuing smoothing effect can be seen so it has
been considered representative for the study of large scale wind power.
Electrical load is characterised by a daily pattern, higher on weekdays than weekends. In
addition to daily cycles, strong temperature dependence can be seen in the Nordic countries.
Wind power has a slightly positive correlation with the load, especially in Denmark.
However, during the winter months, the correlation is practically non existent. The average
wind power production at low temperatures of below –15 oC is somewhat lower than average
in Finland, and these are the incidents of highest load. However, the average wind power
production in Denmark as well as the total Nordic wind power does not experience this kind
of reduction. The average wind power production in times of the highest peak load hours was
near average production in the example years 2000 and 2001.
The need for more flexibility in the electricity system, due to short term variations of wind
power, were estimated with the hourly time series for wind production and load. Net load
variations (load minus wind production) compared to load variations give an estimate for the
needs of the system to react to large scale wind power. The increase in hourly variations was
estimated for Denmark and Finland. An analysis based on only the maximum hourly variation
was found to be very sensitive to the hourly data in question; giving different results for
different years of data, depending on what was the wind power change during the critical
hours of maximum load changes. A probabilistic approach gave estimates for the range of
variations, from the standard deviation (σ) values, taking ±4σ as the range that covers most
variations (99.99 % of all variations are inside this range). The results were that at 5 % wind
power penetration (of gross demand) the increase of variations was 0.5–1 % of installed wind
power capacity, at 10 % penetration 1–2 % and at 15 % penetration1.7–.2.7 % of installed
wind power capacity. The effect of wind power on variations was smaller in Denmark than in
Finland. This is mainly based on the relatively larger load variations in Denmark, absorbing
wind variations. If the Nordic electricity market area would be working without bottlenecks of
transmission, 10 % of wind energy distributed in the area would require extra flexibility of
less than 1 % of installed capacity at 10 % wind penetration (of gross demand). In other
words, 19 000 MW of wind power in the Nordic countries would increase the hourly
76
variations by about 160 MW. However, large scale wind power production could also
increase the bottlenecks.
The estimates of increase in hourly variations do not take into account the fact that the
variations are easier to predict for the load than for wind power production. To estimate the
effect of load and wind forecasts to these analyses, a case for Finland year 2001 load
estimates was run based on the information from previous, year 2000 load data. This analysis
showed that the results above, based on the simple hourly variations from load and wind
power time series, should be increased by 50–100 % depending on the level of wind power
forecast (no forecast versus forecast being 20 % better than not using any). This means that
when producing 10 % of yearly electricity consumption with wind power, the increase in
hourly variations would be 1.5–4 % of the installed wind power, instead of 1–2 % neglecting
the forecasts. More specifically, for Denmark 2000 MW of wind power would increase the
hourly variations by 30–40 MW, for Finland 4000 MW wind power by 120–160 MW and for
the Nordic countries 19 000 MW wind power by 240–320 MW. This can be used as an
estimate for the increase in requirements for load following, or secondary reserve for the
power system due to wind power.
The estimation is based on hourly wind power and load data from two years. The years were
less than average wind years, meaning that the hourly variations are probably underestimated.
However, the smoothing effect of thousands of wind turbines at hundreds of wind farm sites
is also underestimated by the wind power data sets used here. This means that the estimates
for the variations of wind power production are probably still somewhat conservative.
Another basic assumption is that the hourly variations give an estimate of the short term
variations relevant for operating reserve of the power system. Secondary reserve is operated
in 10–15 minutes, and hourly data is used here, as 15 minute data is very limited and would
not allow for a large scale system studies. However, as the wind varies less within an hour
than on hourly basis, using hourly data would not underestimate the effects.
The conclusion of this study is that the hourly variations of large scale wind power will be
seen as an increase in the hourly variations and thus operating reserve requirements of the
power system. The impact will increase the larger share of gross demand is produced by wind
power. At a 10 % wind power penetration level this is estimated as 1.5–4 % of installed wind
capacity, taking into account that load variations are more predictable than wind power
variations.
The cost of this increase in operating reserves, as well as the electricity market studies,
focusing on longer term variations of wind power, are subjects for future work.
77
References
Alm, L K, Tallhaug, L, 1993. Effektbidrag fra vindkraft (Wind power contribution to power
adequacy). IFE report IFE/KR/F-93/132, Halden, Norway (in Norwegian).
Carlstedt, N-E, 2003. Driftuppföljning av vindkraftverk, årsrapport 2002. Elforsk report
03:12, STEM report 12:2003. Stockholm, Sweden. http://www.elforsk.se/varme/varmvind.html
CER/OFREG NI, 2003. Impacts of increased levels of wind penetration on the electricity
systems of the republic of Ireland and Northern Ireland: Final report. A report commissioned
by Commission for Energy Regulation in Republic of Ireland and OFREG Northern Ireland.
Available at http://www.cer.ie/cerdocs/cer03024.pdf
Elkraft, 2003. http:/www.elkraft.dk/ 1.9.2003
Eltra, 2003. http:/www.eltra.dk/ 1.9.2003
Ensslin, C, Hoppe-Kilpper, M, Rohrig, K, 2000. Wind power integration in power plant
scheduling schemes. Proceedings of EWEC Special Topic conference, Kassel, 25-27th
September, 2000.
Eriksen, P B, Pedersen, J, Parbo, H, 2002. Challenges of Large-Scale Integration of
Distributed Generation into Eltra’s System. Proceedings of 2nd International Symposium on
Distributed Generation: Power System and Market Aspects, Stockholm, 2 - 4 October 2002.
Ernst, B, 1999. Analysis of wind power ancillary services characteristics with German 250
MW wind data. 38 pp.; NREL Report No. TP-500-26969 available at
<http://www.nrel.gov/publications/>
Fingrid, 2003. http://www.fingrid.fi/engl/palvelut/palvelut_vaakajako.html , 5.12.2003
Focken, U, Lange, M, Waldl, H-P, 2001. Previento - A Wind Power Prediction System with
an Innovative Upscaling Algorithm. Proceedings of EWEC’01, 2nd-6th July, 2001,
Copenhagen.
Giebel, G, 2000. Equalizing effects of the wind energy production in Northern Europe
determined from Reanalysis data. Risö-R-1182(EN), Roskilde, available at
http://www.risoe.dk/rispubl/index.htm
Giebel, G, 2001. On the Benefits of Distributed Generation of Wind Energy in Europe.
Fortshr.-Ber. VDI Reihe 6 Nr 444. Düsseldorf, VDI Verlag, 2001. 116 p. ISBN 3-18-3444062, ISSN 0178-9414, available at http://www.drgiebel.de/thesis.htm
Giebel, G, Karionakis, G, Brownsword, R, 2003. The state-of-the-art in short-term prediction
of wind power from a Danish perspective. In proceedings of the 4th International Workshop
on Large-Scale Integration of Wind Power for Offshore Wind Farms. Billund, Denmark,
October, 20-21, 2003. KTH, Sweden, 2003.
78
Grubb, M J, 1991. The integration of renewable energy sources. Energy Policy September
1991.
Hirst, E, 2002. Integrating wind output with bulk power operations and wholesale electricity
markets. Wind Energy (5) 19-36, 2002.
Hirvonen, R, 2000. Material for course S-18.113 Sähköenergiajärjestelmät, Helsinki
University of Technology, Power Systems laboratory (in Finnish).
Holttinen, H, Peltola, E, Koreneff, G, 1996. Seasonal and long-term variations of wind power
production in Finland (in Finnish). 42 p. + app 9 p. VTT - Research Notes; 1800. Espoo.
ISBN 951-38-4995-3.
Holttinen, H, Nielsen, T S, Giebel, G, 2002. Wind energy in the liberalised market - forecast
errors in a day-ahead market compared to a more flexible market mechanism. Proceedings of
2nd International Symposium on Distributed Generation: Power System and Market Aspects,
Stockholm,
2
4
October
2002.
Session
6.
13
p.
http://www.energia.fi/page.asp?Section=1354&Item=3953
Holttinen, H, 2002. The impacts of hourly wind variations in the needs for system flexibility
for large scale wind power production in the Nordic countries. Global Wind Power
Conference. Proceedings. Paris 2 - 5 April 2002. 5 p. CD-ROM, European Wind Energy
Association.
Holttinen, H, Pedersen, J, 2003. The effect of large scale wind power to a thermal system
operation. In proceedings of the 4th International Workshop on Large-Scale Integration of
Wind Power for Offshore Wind Farms. Billund, Denmark, October, 20-21, 2003. KTH,
Sweden, 2003.
Holttinen, H, Hirvonen, R, 2004. Power system requirements of wind power. Chapter 10 in
Wind Energy in Power Systems, ed. T.Ackermann, Wiley, in print.
Hurley, T, Watson, R, 2002. An assessment of the expected variability and load following
capability of a large penetration of wind power in Ireland. Proceedings of Global Wind Power
Conference GWPC´02 Paris.
ISET, 2002. Wind energy report Germany. Institut für Solare Energieversorgungstechnik
ISET, Kassel, Germany, 2002.
Kirby, B, Hirst, E, 2000. Customer-specific metrics for the regulation and load following
ancillary services. Oak Ridge National Laboratory.
Koreneff, G, Seppälä, A, Lehtonen, M, Kekkonen, V, Laitinen, E, Häkli, J, Antila, E, 1998.
Electricity spot price forecasting as a part of energy management in de-regulated power
market. Proceedings of EMPD '98. 1998 International Conference on Energy Management
and Power Delivery. Singapore, 3 - 5 March 1998. Vol. 1. IEEE. pp 223 – 228.
Koreneff, G, Kekkonen, V, 2000. Distribution Energy Management (DEM) Systems and
Management of Commercial Balance. NORDAC 2000 -Nordic Distribution Automation
79
Conference. Stjördal 22 - 23 May 2000. Trondheim, DEFU in Denmark; VTT Energy;
SINTEF Energy Research; Elforsk, 2000. 10 p.
Kristoffersen, J R, Christiansen, P, Hedevang, A, 2002. The wind farm main controller and
the remote control system in the Horns Rev offshore wind farm. Proceedings of Global Wind
Power Conference GWPC´02 Paris.
Laakso, T, 2003. The statistics of wind power in Finland. Yearly report 2002 (in Finnish).
http://www.vtt.fi/pro/pro2/tuulitilastot/windstat.htm
Landberg, L, 1997. The availability and variability of the European wind resource. Int J Solar
Energy 18 pp. 313-320 (1997).
Lumituuli, 2003 http://www.lumituuli.fi 1.9.2003
Lund, H., Münster, E., 2003. Management of surplus electricity production from a fluctuating
renewable energy source. Applied Energy 76 (2003) 65-74.
Milborrow, D, 2001. Penalties for intermittent sources of energy. Submission to Energy
policy review, September 2001. Available at http://www.pm.gov.uk/output/Page77.asp or
directly http://www.number10.gov.uk/output/Page3703.asp
Milligan, M, Miller, A H, Chapman, F, 1995. Estimating the economic value of wind
forecasting to utilities. Proceedings of Windpower’95, Washington D.C, March 27-30, 1995.
A report NREL/TP-441-7803 available at http://www.nrel.gov/publications/
Milligan, M, 2000. Modelling Utility-Scale Wind Power Plants. Part 2: Capacity Credit. Wind
Energy 3 2000 pp 167-206.
Milligan, M, Factor, T, 2000. Optimizing the geographic distribution of wind plants in Iowa
for maximum economic benefit and reliability. Wind Engineering. Vol. 24(4), 2000. pp 271–
290.
Milligan, M, Porter, K, Parsons, B & Caldwell, J, 2002. Wind energy and power system
operations: a survey of current research and regulatory actions. Electricity Journal. Vol. 15(2),
March
2002;
pp.
56-67;
NREL
Report
No.
31381
available
at
http://www.nrel.gov/publications/
Milligan, M, 2003. Wind power plants and system operation in the hourly time domain.
Proceedings of Windpower 2003 conference, May 18–21, 2003 Austin, Texas, USA.
NREL/CP-500-33955 available at http://www.nrel.gov/publications/
Naturlig Energi, 2003. http://www.naturlig-energi.dk
Nordel, 2000. Ikke-regulerbar produktion i Nordel-systemet. Nordels netgruppe, 2000.
Nordel, 2003. Statistics of power production and consumption in the Nordic countries. Yearly
statistics 2001. In http://www.nordel.org/Content/Default.asp?PageID=157
80
Nordpool, 2003. http://www.nordpool.com/products/index.html , 5.12.2003
Nørgaard, P, 2003. A multi-turbine power curve. Note for task 2 of EU funded research
project WILMAR: Wind power Integration in Liberalised electricity markets
http://www.wilmar.risoe.dk/
Parsons, B K, Wan, Y, Kirby, B, 2001. Wind farm power fluctuations, ancillary services, and
system operating impact analysis activities in the United States. Proceedings of EWEC’01,
2nd-6th July, 2001, Copenhagen; NREL Report No. CP-500-30547 available at
http://www.nrel.gov/publications/
Peltola, E, Petäjä, J, 1990. Modelling of wind power in the Finnish power supply system. In
Proceedings of European Wind Energy Conference EWEC’90, 10-14th September, Madrid,
Spain.
Persaud, S, Fox, B, Flynn, D, 2000. Modelling the impact of wind power fluctuations on the
load following capability of an isolated thermal power system. Wind Engineering (24) no 6.
pp 399-415.
Poore, R Z, Randall, G, 2001. Characterizing and predicting ten minute and hourly
fluctuations in wind power plant output to support integrating wind energy into a utility
system. Proceedings of AWEA Windpower’01 conference.
Pryor, S C, Barthelmie, R J, 2001. Persistence of offshore winds: implications for power
quality. Proceedings of EWEC’01, 2nd-6th July, 2001, Copenhagen.
SMHI, 2003. Wind speed measurement time series as a courtesy of Hans Bergström, SMHI.
Seppälä, A, 1996. Load research and load estimation in electricity distribution. VTT
Publications 289, 118 p. + app. 19 p. Espoo, Finland: ISBN 951-38-4947-3
Söder, L, 1999. Wind energy impact on the energy reliability of a hydro-thermal power
system in a deregulated market. In Proceedings of Power Systems Computation Conference,
June28-July 2, 1999, Trondheim, Norway.
Tammelin, B, Nurmi, J, 2001. An assessment of variability of wind power production in
Finland. In Proceedings of European Wind Energy Conference EWEC’01, 2-6th July, 2001,
Copenhagen, Denmark.
van Wijk, A J M, 1990. Modelling wind power production in the Netherlands. Wind
Engineering Vol. 14 No. 2, 1990.
van Zuylen, E J, Ramaekers, L A M, van Wijk, A J M, Verschelling, J A, 1994. Wind power
fluctuations on a national scale. In Proceedings of European Wind Energy Conference
EWEC’96, 20-24th May, 1996, Göteborg, Sweden.
Windpower Monthly, 2003. Windicator, Windpower Monthly magazine, April 2003.
81
APPENDIX 1
1
Denmark, average 22 %
100 %
2000
80 %
60 %
40 %
20 %
0%
1
741
1481
2221
2961
3701
4441
5181
5921
6661
7401
8141
Denmark, average 22 %
100 %
2001
80 %
60 %
40 %
20 %
0%
1
741
1481
2221
2961
3701
4441
5181
5921
6661
7401
8141
Finland, average 23 %
100 %
2000
80 %
60 %
40 %
20 %
0%
1
741
1481
2221
2961
3701
4441
5181
5921
6661
7401
8141
APPENDIX 1
Finland, average 23 %
100 %
2001
80 %
60 %
40 %
20 %
0%
1
741
1481
2221
2961
3701
4441
5181
5921
6661
7401
8141
Figure 1. Hourly wind power production time series for Denmark and Finland, example years 2000 and 2001.
The production values as % of capacity (y-axis). On x-axis, the hour of the year is marked at 740 hour (about
one month) intervals. Average production for the whole period 2000-2001 is denoted in the title.
Norw ay, average 33 %
100 %
2000
80 %
60 %
40 %
20 %
0%
1
741
1481
2221
2961
3701
4441
5181
5921
6661
7401
8141
Norw ay, average 33 %
100 %
2001
80 %
60 %
40 %
20 %
0%
1
741
1481
2221
2961
3701
4441
5181
5921
6661
7401
8141
2
APPENDIX 1
3
Sw eden, average 24 %
100 %
2000
80 %
60 %
40 %
20 %
0%
1
741
1481
2221
2961
3701
4441
5181
5921
6661
7401
5921
6661
7401
8141
Sw eden, average 24 %
100 %
2001
80 %
60 %
40 %
20 %
0%
1
741
1481
2221
2961
3701
4441
5181
8141
Figure 2.Hourly wind power production time series for Norway and Sweden, example years 2000 and 2001. The
production values as % of capacity (y-axis). On x-axis, the hour of the year is marked at 740 hour (about a
month) intervals. Average production for the whole period 2000-2001 is denoted in the title.
APPENDIX 1
4
Nordic, average 26 %
100 %
2000
80 %
60 %
40 %
20 %
0%
1
741
1481
2221
2961
3701
4441
5181
5921
6661
7401
8141
5921
6661
7401
8141
Nordic, average 26 %
100 %
2001
80 %
60 %
40 %
20 %
0%
1
741
1481
2221
2961
3701
4441
5181
Figure 3. Hourly wind power production time series for example years 2000 and 2001, assuming same capacity
in all 4 countries Sweden, Norway, Denmark and Finland. The production values as % of capacity (y-axis). On
x-axis, the hour of the year is marked at 740 hour (about a month) intervals. Average production for the whole
period 2000-2001 is denoted in the title.
APPENDIX 2 1
hourly variation (% cap)
Denmark
25 %
2000
15 %
5%
-5 %
-15 %
-25 %
1
741
1481
2221
2961
3701
4441
5181
5921
6661
7401
8141
5181
5921
6661
7401
8141
5181
5921
6661
7401
8141
5181
5921
6661
7401
8141
hourly variation (% cap)
Denmark
25 %
2001
15 %
5%
-5 %
-15 %
-25 %
1
741
1481
2221
2961
3701
4441
hourly variation (% cap)
Finland
25 %
2000
15 %
5%
-5 %
-15 %
-25 %
hourly variation (% cap)
1
741
1481
2221
2961
3701
4441
Finland
25 %
2001
15 %
5%
-5 %
-15 %
-25 %
1
741
1481
2221
2961
3701
4441
hour
Figure 1.Time series for hourly variations of wind power production for Denmark and Finland, example years
2000 and 2001. Positive means increasing and negative decreasing wind power production.
APPENDIX 2 2
hourly variation (% cap)
Sweden
25 %
2000
15 %
5%
-5 %
-15 %
-25 %
1
741
1481
2221
2961
3701
4441
5181
5921
6661
7401
8141
5181
5921
6661
7401
8141
5181
5921
6661
7401
8141
5181
5921
6661
7401
8141
hourly variation (% cap)
Sweden
25 %
2001
15 %
5%
-5 %
-15 %
-25 %
1
741
1481
2221
2961
3701
4441
hourly variation (% cap)
Norway
25 %
2000
15 %
5%
-5 %
-15 %
-25 %
1
741
1481
2221
2961
3701
4441
hourly variation (% cap)
Norway
25 %
2001
15 %
5%
-5 %
-15 %
-25 %
1
741
1481
2221
2961
3701
4441
hour
Figure 2.Time series for hourly variations of wind power production for Sweden and Norway, example years
2000 and 2001. Positive means increasing and negative decreasing wind power production.
APPENDIX 3 1
Denmark 2000
7000
hourly load (MW)
6000
5000
4000
3000
2000
1000
0
1
721
1441
2161
2881
3601
4321
5041
5761
6481
7201
7921
8641
5041
5761
6481
7201
7921
8641
5041
5761
6481
7201
7921
8641
Finland 2000
14000
hourly load (MW)
12000
10000
8000
6000
4000
2000
0
1
721
1441
2161
2881
3601
4321
Norway 2000
25000
hourly load (MW)
20000
15000
10000
5000
0
1
721
1441
2161
2881
3601
4321
hour
Figure 1.Hourly electricity consumption, the load, for Denmark, Finland and Norway, example year 2000.
APPENDIX 3 2
Sweden 2000
30000
hourly load (MW)
25000
20000
15000
10000
5000
0
1
721
1441
2161
2881
3601
4321
5041
5761
6481
7201
7921
8641
5041
5761
6481
7201
7921
8641
Nordic 2000
70000
hourly load (MW)
60000
50000
40000
30000
20000
10000
0
1
721
1441
2161
2881
3601
4321
hour
Figure 2. Hourly electricity consumption, the load, for Norway and the combined load for the 4 Nordic countries
Denmark, Finland, Norway and Sweden , example year 2000.
APPENDIX 4 1
Denmark 2000
hourly load variation (MW)
1500
1000
500
0
-500
-1000
-1500
1
741
1481
2221
2961
3701
4441
5181
5921
6661
7401
8141
5181
5921
6661
7401
8141
5181
5921
6661
7401
8141
Finland 2000
hourly load variation (MW)
1500
1000
500
0
-500
-1000
-1500
1
741
1481
2221
2961
3701
4441
Nordic 2000
7000
hourly load variation (MW)
5000
3000
1000
-1000
-3000
-5000
-7000
1
741
1481
2221
2961
3701
4441
Figure 1. Time series of hourly load variations, for Denmark, Finland and the combined load for the 4 Nordic
countries Denmark, Finland, Norway and Sweden , example year 2000.
APPENDIX 4 2
Denmark 2001
hourly load variation (MW)
1500
1000
500
0
-500
-1000
-1500
1
741
1481
2221
2961
3701
4441
5181
5921
6661
7401
8141
5181
5921
6661
7401
8141
5181
5921
6661
7401
8141
Finland 2001
hourly load variation (MW)
1500
1000
500
0
-500
-1000
-1500
1
741
1481
2221
2961
3701
4441
Nordic 2001
7000
hourly load variation (MW)
5000
3000
1000
-1000
-3000
-5000
-7000
1
741
1481
2221
2961
3701
4441
Figure 2. Time series of hourly load variations, for Denmark, Finland and the combined load for the 4 Nordic
countries Denmark, Finland, Norway and Sweden , example year 2001.