Demand Optimization. Jason W Black Nov 2, 2010 University of Notre Dame. December 3, 2010

Transcription

1 Demand Optimization Jason W Black (blackj@ge.com) Nov 2, 2010 University of Notre Dame 1

2 Background Demand response (DR) programs are designed to reduce peak demand by providing customers incentives to use E le c tricity energy at different (lower cost) times kw h There are two classes of DR programs: P e a k e rs C o m b in e d c y c le N u cle a r 4 P M D a ily 1. Direct load controls- Allow utility to remotely control loads via on/off switches or thermostat adjustments 2. Dynamic Pricing Allows adjustment of retail prices during peak hours according to system conditions (peak load or price) C o a l 2

3 Demand Resource Capabilities Shift Peaks Increase Reliability E le ctricity kw h P e a kers C o m b in e d cycle C o a l N u c le a r 4 P M D a ily Demand Resource as a Virtual Peaking Generator Demand Resource responds to contingencies Fill Gaps Smooth Variations PV + Demand Response PV DR PV+DR Demand Resource to complement high penetration of intermittent generation Demand Resource for Frequency Regulation H our

4 Consumer Objectives Reduce Costs Maintain Comfort Minimize utility bills without accepting exposure to volatility Do not compromise on essential services, temperature, etc. Be Green Improve Service Reduce Environmental impacts includes integration of PEV, renewables Reduce outage frequency and duration, improve power quality 4

5 Demand Resource Opportunities Nearly 75% of residential electricity demand can be controlled with little/no impact on service Elect ronics 7.3% Other 10.4% Fixed HVAC 29.4% Typical US Household Light ing 8.8% Cooking 7.0% Dishwasher 2.5% Washer/ Dryer 6.7% 26% Deferrable 16% Storable 57% Refrigerat ion 17.2% Pool 1.6% Hot Water 9.1% 5

6 Emerging Challenges with Demand Response Rebound Effect Creates a new peak and marginalizes benefits Reliable Response Limited forecasting capability ~1 KW Drop ~1.5 KW Increase C P P w /PC T C P P no P CT C ontrol (no C P P ) Coincident Shifting Magnifies load changes, aggravated by automation and fixed time periods Every 1,000 customers participating can create > 1 MW jump/drop in load nearly instantaneously Requires advanced solutions 6

7 Demand Response: Approach GE s Product 7

8 DR1000 Demand Response: Integrated Approach Solution Architecture 8

9 DR1000 Demand Response: Operations Approach Module All the critical applications to manage and maximize resources for demand response events 9

10 Demand Response: Approach Research 10

11 Current Research in Demand Optimization 1. Response Estimation - Develop an estimation tool to provide utilities with the highest fidelity estimates of response magnitude (MW) and duration (minutes/hours) for a variety of DR programs (DLC, CPP, etc.). 2. Advanced Aggregation - Develop the algorithms necessary to dynamically aggregate load resources (customers) at any node on the grid and to select/schedule customers to maximize economic benefit. 3. Demand Dispatch - Develop enhanced demand dispatch capability including Direct Load Control, Critical Peak Pricing and Peak Time Rebate programs. 4. Contingency Response - Evaluate distribution level contingencies that can be mitigated by the use of demand response. 5. PHEV Integration (Smart charging) - Determine the optimal charging schemes for Plug in Electric Vehicles (PEVs). 11

12 Demand Response: Aggregation/Disaggregation Approach Challenge Current Demand Response Solutions use static customer groupings with all or one dispatch of each group for particular DR events. They are unable to dynamically aggregate consumers based on the electric network model, to select only the minimum set of customers, and to target specific network nodes. Solution This tool will include algorithms necessary to dynamically aggregate load resources (customers) at any node on the grid and to select/schedule customers to maximize economic benefit. It will enable contingency response at the distribution level and vastly enhance the overall capabilities of DR programs for utilities.

13 Initial status Alarm on Distribution Feeder Need to selectively shed load in affected area to relieve overload Avoid need for rolling blackouts 13

14 Load availability Determine available load shed at affected node 14

15 Geospatial view of selected homes Identified Homes: Shed verified Failures/ Overrides Noted 15

16 Update system status Alarm Cleared! 16

17 Dynamic Aggregation and Disaggregation Connected Critical Node Not Connected Aggregation: - Aggregate by: node/location, customer type, device type, etc. - Satisfy power flow constraints - Utilize Network Model - Enable targeted demand response Disaggregation: - Choose only the # of houses necessary - Include duration and rebound - Do not over shed/ create rebound peak - Balance across phases and feeders 17

18 Customer Selection and Scheduling (Knapsack) Rebound Effect* Selections based on: Shed Amount Premise 4 Premise 2 Premise 5 Premise 6 Premise 3 Premise 7 Cost Load Available Response Duration Rebound Effect Contracts Equitable Rotation Premise 1 Duration *Additional Reduction necessary over time to account for rebound effect Determine least cost schedule (with lowest customer impact) 18

19 Demand Response: Dispatch Approach Challenge Develop enhanced demand dispatch capability for Critical Peak Pricing and Peak Time Rebate programs. Solution There are 4 major capabilities being developed: Determine optimal event window (start and end times) Determine optimal group selection for CPP. Enable a selection of the subset of population for each event. Determine action timing for peak time rebate (PTR). This will evaluate when utilities should invoke a PTR event, based on the cost of the incentive payments to customers, the market/generation costs, and estimated responses. Enable optimal scheduling of groups. Combination of 1 and 2 above, where various groups can be assigned different windows within a day to create an optimal overall response.

20 PHEV Integration Distribution Transformer Load Without smart charging With smart charging Load shifted Available Capacity Fixed PEV Load + Normal Load Normal (Non-PEV) Load Objective: Schedule Smart Charging Load. Subject to: Desired Transformer power consumption Controllable PEV loads and their current SOC, charging window, charging mode capability, and PEV charging profiles. Transform cooling down constraints. Equitable charging (e.g., rotational 1 hour increment) while maintaining priority based upon time constraints and charging rates. 20

21 Demand Response: Approach Demo 2 Real Options 21

22 Problem DR programs typically have constraints on the number of days/hours that a high price event can be called Utilities need a method to optimally allocate the scarce events across a season/year Optimal allocation would match the highest wholesale price days/hours with DR events 22

23 Example California Critical Peak Pricing (CPP) events per year /KWh Hourly Prices Normal Day CPP Day 2. Must declare event 1 day prior 3. Limited to June 1 to October Fixed Prices and Time periods (2-7 pm) Available Information Day Ahead Wholesale Prices** Weather Forecast (n days) Historical price and weather data Load Forecast (based on weather) **May not be available by the time event has to be called 23

24 Current Approach Temperature heuristic Since high temperatures drive peak load conditions, utilities currently use a temperature trigger to call CPP events Set and initial temperature threshold and invoke an event when that threshold is exceeded Update threshold every 2 weeks based on days remaining, number of events used, and weather trends (expert adjustment) But Wholesale prices are highly volatile, with relatively low correlation to temperature or load 24

25 Price-Temperature Correlation (CA-ISO) Low correlation holds across multiple US electricity markets many sources of volatility 25

26 Our Approach Options Valuation 1. Determine temperature and price distributions 2. Adjust distributions for known information (e.g. weather forecast) 3. Use dynamic programming approach to determine option value (opportunity cost) of each remaining event Data: Decision variables: Dynamic Programming Recursion: 26

27 Case Study Daily Threshold Today Date 9/1/2009 Events Remaining 4 Threshold 94.2 If Event Called Tomorrow Date 9/2/2009 Events Remaining 3 Threshold 95.0 If Event Not Called Date 9/2/2009 Events Remaining 4 Threshold

28 Temperature Approach Jun 8-Jun 15-Jun 22-Jun 29-Jun 6-Jul 13-Jul 20-Jul 27-Jul 3-Aug 10-Aug 17-Aug 24-Aug 31-Aug 7-Sep 14-Sep 21-Sep 28-Sep Average Temperature on Selected days Utility Heuristic Method: 92.4 GE Option Value Method: 94.3 Option Threshold Utility Threshold Option Event Utility Event

29 Price and Temperature Approaches 11% improvement from utility heuristic to option-approach 15% improvement if use price and temperature together Model performs better in high-price years than low-price years (events called at end of period) Better forecasting techniques are critical 29