Adaptive Load Management Adaptive Load Management Jhi‐Young (J.Y.) Joo and Marija Ilić Electric Energy Systems Group Electrical and Computer Engineering Carnegie Mellon University This work is supported in part by Bosch. Outline Adaptive load management (ALM) Adaptive load management (ALM) Future electric energy systems DYMONDS and demand response demand response Motivation and main ideas Schematics – Schematics – information flow information flow Cost analysis Future work and open questions Future work and open questions 2 Future electric energy systems Future electric energy systems Distributed optimization / decision‐making Flexible & dynamic system components Higher uncertainty Artwork by Andrew Hsu, EESG 3 Why demand response? Why demand response? Change in supply attributes g pp y Distributed generation resources Higher uncertainty, intermittency Requires quick response of the system (higher ramping Requires quick response of the system (higher ramping rates) Higher cost of additional capacity Higher cost of additional capacity Need for load shifting End‐users’ needs Energy efficiency and lower energy cost 4 Elastic demand with wind generation [2] Elastic demand with wind generation elastic demand wind System setup System setup Demand responsive to hourly real‐time pricing of the system Æ demand curves Price reflecting the true Price reflecting the true cost of the supply including cost of ramping rates i t Æ look‐ahead dispatch IEEE 24‐bus Reliability Test System‐1996 5 Elastic demand with wind generation [2] Elastic demand with wind generation If demand can respond p rapidly to the system price, it can compensate for the compensate for the high intermittency of renewable resources. What do we need in terms of system operation? 6 The main ideas of ALM The main ideas of ALM System’s perspective: What is the right information required to b l balance the system with demand response? th t ith d d ? Where do we see the value of DR in the system balancing mechanism? End‐users’ End‐users perspective: Reflecting various end‐users perspective: Reflecting various end‐users’ needs needs and and preferences End‐users’ willingness to pay for energy service Load aggregators’ role Physical energy mediator between system/market and end‐users Information exchange junction/processor between market and end‐ users Value of aggregating different resources and risk management 7 Previous demand response scheme: Direct load control One‐way flow of information Load management conducted by utilities Top‐down control Exclusive contracts between supply and demand Direct load control Regardless of end‐users’ preferences No access to market No access to market information for end‐users End‐users’ information invisible to system 8 Information flow of ALM Tertiary layer Bid function b(λ) Market price Market price λ Secondary layer Load aggregator I Demand function b(λI) Load aggregator II Load aggregator III End‐user rate λI Primary layer End‐user 9 Multi‐layered ALM – end‐users to LAs Multi layered ALM end users to LAs Tertiary level Bid function y(λ) Market price Market price λ Secondary level Load aggregator I Demand function x(λI) User rate λI Primary level End‐user Load aggregator II Load aggregator III Multi‐layered Multi layered ALM ALM – end end‐users users to LAs to LAs Obtaining individual demand function subject to temperature comfort level ∑ [λ k0 + N min {xi } k =k 0 where LA {( [k ] ⋅ xi [k ] + Ti [k ] − Ti ) + (T [k ] − T ) }] max 2 min 2 i i Ti [k + 1] = AiTi [k ] + Bi xi [k ] subject to Ti min ≤ Ti [k ] ≤ Ti max subject to for all k for all xi [k ] Obtain different optimal energy usage ’s for different temperature setpoint λ [k ]’ss to infer demand different temperature setpoint to infer demand functions 11 Demand function Demand function Mathematical model of price‐responsive loads Integrated into the system optimization Includes information of eend‐users’ utility (benefit) d use s u y (be e ) Function of end‐users’ willingness‐to‐pay willingness to pay With respect to electricity demand quantity cost/price ($/MWh) λ demand (function) supply λ* P* d (PD ) = aPD + b quantity (MWh) Demand function (cont Demand function (cont’d) d) How to obtain from individual temperature control Calculate optimal energy usage by hours with a given electricity price Perturb the given price by a certain percentage (e.g. ±20%) Perturb the given price by a certain percentage (e g ±20%) and re‐calculate optimal energy usage with new prices Curve‐fit price‐demand quantity pairs to identify the parameters of a demand function parameters of a demand function Unit price (cents/kWh) Demand (kW) Cost analysis of ALM Cost analysis of ALM Problem setup ob e setup 10 end‐users with different temperature preferences Optimizing energy usage over 24 O i ii 24 hours Hourly‐varying electricity price given (real‐time pricing) Outdoor weather temperature given price (¢/kWh) p Hourly real-time pricing rates 14 12 10 8 6 4 2 0 1 3 5 7 9 11 13 hour 15 17 19 21 23 End­ End user index Temperature setpoints (⁰F) 1 68 75 2 70 77 3 72 75 4 74 79 5 75 80 6 64 75 7 63 77 8 72 79 9 72 77 10 73 81 14 Cost analysis of ALM (cont Cost analysis of ALM (cont’d) d) Demand functions of end‐user Demand functions of end user #1 #1 hour 14 50 40 30 20 10 0 hour 16 WTP (¢/kkWh) 50 40 30 20 10 0 hour 15 WTP (¢/kkWh) WTP (¢//kWh) 50 40 30 20 10 0 50 40 30 20 10 0 0 0.1 0.2 0.3 0.4 0 0.1 0.2 0.3 0.4 0 0.1 0.2 0.3 0.4 0 0.1 0.2 0.3 0.4 energy usage (kWh) energy usage (kWh) energy usage (kWh) gy g ( ) energy usage (kWh) gy g ( ) hour 17 hour 18 50 40 30 20 10 0 hour 20 WTP (¢/kWh) 50 40 30 20 10 0 hour 19 WTP (¢/kkWh) 50 40 30 20 10 0 WTP (¢/kkWh) WTP (¢/kkWh) WTP (¢/kkWh) hour 13 50 40 30 20 10 0 0 0.1 0.2 0.3 0.4 0 0.1 0.2 0.3 0.4 0 0.1 0.2 0.3 0.4 0 0.1 0.2 0.3 0.4 energy usage (kWh) energy usage (kWh) energy usage (kWh) energy usage (kWh) ( ) energy usage (kWh) 15 Cost analysis of ALM (cont Cost analysis of ALM (cont’d) d) Cost savings from “price‐ conscious” optimization End‐users' End users energy costs energy costs (hourly RTP) Current practice : end‐users little concerned about hourly prices b th l i Intelligent optimization with ALM : optimization w r t time‐ : optimization w.r.t. time varying price saves costs 32% of cost savings (as a whole) in this case temp+price optimization p p p 20 Cost ($/mo onth) Introduction of real‐time pricing may end up in energy cost rise for some end‐users d temp. only optimization 15 10 5 0 1 2 3 4 5 6 7 8 9 10 End‐user index 16 Cost analysis of ALM (cont Cost analysis of ALM (cont’d) d) Does real‐time pricing p g save costs for all end‐ users? fixed rate 1 fixed rate 2 2.5 2 cost (($/month) Not Not necessarily necessarily Different end‐users’ load profiles have different values values. Need to aggregate different end‐users’ loads in order to distribute cost in order to distribute cost or benefit properly and maximize profit Energy cost comparison with RTP 1.5 1 0.5 0 ‐0.5 1 2 3 4 5 6 7 8 9 10 ‐1 ‐1.5 end‐user end user index index Fixed rate 1: 7.24 ¢/kWh Fixed rate 2: 7.84 ¢/kWh 17 Open questions Open questions Who benefits from the cost savings? It depends on Load aggregator’s portfolio in various energy markets d ’ f l k Long‐term contracts, day‐ahead/real‐time markets, ancillary service markets, etc. Load aggregator’s aggregation of available resources Elastic demand, distributed energy resources (e.g. small‐scale renewables, electric vehicles) hi l ) Load aggregator’s contracts with end‐users Real‐time pricing, time‐of‐use pricing, flat rates, interruptible loads, etc. What is the smartest way to get end‐users’ demand preferences? Wh t i th t t t t d ’d d f ? Temperature settings control Price response analysis Æ hands‐on experiment on end‐users Typical load profile analysis yp p y Behavior analysis on energy usage and so on… 18 References References 1) J.Y. Joo and M. Ilić, A Multi‐Layered Adaptive Load Management (ALM) System, IEEE PES Transmission and Distribution Conference, April 2010, accepted 2) L. Xie and M. Ilić, Model Predictive Economic/Environmental Dispatch of Power Systems with Intermittent Resources, IEEE PES General Meeting, July 2009 3) M. M. Ilić, L. Xie, and J.Y. Joo, Efficient Coordination of Wind Ilić, L. Xie, and J.Y. Joo, Efficient Coordination of Wind Power and Price‐Responsive Demand Part I: Theoretical Foundations, Part II: Case Studies, IEEE Transactions on Power Systems, under review, y , , Mar 2010 19 20 Cost analysis of ALM (cont Cost analysis of ALM (cont’d) d) ALM applied to a distribution network app ed to a d st but o et o Source: Judith Cardell, Control Strategies and Dynamic Pricing g y g for Small Scale Distributed Generation in a Deregulated Market, PhD Thesis, MIT, 1998 Economic dispatch with demand functions n min ∑ {Ci (Pi ) − Bi (Pi )} P i i =1 subject to Fi , j ≤ Fi , j max 0 ≤ Pi ≤ Pi ,max for ∀i, j 21 Cost analysis of ALM (cont Cost analysis of ALM (cont’d) d) ED with and without demand functions t a d t out de a d u ct o s Without demand functions Inelastic demand : demand does NOT change w.r.t. price d dd NOT h t i With demand functions Elastic demand : demand changes w.r.t. price Æ demand likely to be lower when price higher Cost Cost savings estimate in hour 15 (when demand was savings estimate in hour 15 (when demand was highly elastic) : for 13,000 end‐users, $65.12 savings with ALM 22