Dynamic Testing of Wholesale Power Market Designs: Presenter:

Transcription

1 Dynamic Testing of Wholesale Power Market Designs: An Agent-Based Computational Approach Presenter: Leigh Tesfatsion Professor of Economics and Mathematics Department of Economics Iowa State University Ames, Iowa

2 Outline What is Agent-based Computational Economics (ACE)? ACE and Market Design A Real-World Market Design Project: Construction of an open-source computational laboratory to test the economic and physical reliability of wholesale power market designs Empirical Validation: An Iterative Participatory Modeling Approach 2

3 What is ACE? Constructive approach to the study of decentralized market processes ACE = Computational study of economies as dynamic systems of interacting agents Agent = Bundled methods and data representing an individual, social, biological, or physical entity in a computationally constructed world 3

4 ACE Modeling: Culture Dish Analogy Modeler constructs a virtual economic world populated by various agent types Modeler sets initial world conditions Modeler then steps back to observe how the world develops over time (no further intervention permitted) World events are driven by agent interactions (dynamic completeness) 4

5 Key Characteristics of ACE Models Agents are encapsulated software entities capable (in various degrees) of Adaptation to environmental conditions Social communication with other agents Goal-directed learning Autonomy (self-activation and self-determinism based on private internal processes) Agents can be situated in realistically rendered problem environments Behaviour and interaction patterns can develop/evolve over time 5

6 ACE and Market Design Key Issue: Does a proposed or actual market design ensure efficient, fair, and orderly market outcomes over time despite repeated attempts by traders to game the design for their own personal advantage? ACE Approach: Construct an agent-based world capturing salient aspects of the market design. Introduce self-interested traders with learning capabilities. Let the world evolve. Observe/evaluate resulting market outcomes. 6

7 A Real-World Market Design Project Dynamic Testing of Wholesale Power Market Designs: An Agent-Based Computational Approach Joint research by Junjie Sun (Economics Ph.D. Candidate, ISU) Leigh Tesfatsion (Professor of Economics and Mathematics, ISU) funded in part by the National Science Foundation Work in Progress 7

8 Wholesale Power Market Platform - WPMP Note: U.S. Federal Energy Regulatory Commission (FERC), White Paper (April 2003), proposed WPMP design for common adoption by all U.S. wholesale power markets Policy Objectives: Customer-based competitive wholesale power markets providing reliable service Fair/open access to the transmission grid at reasonable prices Good price signals to encourage appropriate investment in new generation and transmission Market power oversight and mitigation 8

9 FERC s WPMP Design Adopted? Mid-Atlantic States (PJM) implement similar plan (1998) New York (NYISO) implements similar plan (1999) New England (ISO-NE) implements similar plan (2003) California (CAISO) files to adopt similar plan (2003) Midwest (MISO) files to adopt similar plan (7/2003), withdraws filing (10/2003), refiles (3/2004), and then adopts design (4/2005) Opposition from states in Southeast and Northwest 9

10 Existing and Proposed ISO/RTO Regions 10

11 Why Resistance to FERC s WPMP Proposal? Stakeholders in Midwest Key cited problem: Lack of sufficient reliability testing. Stakeholders in Southeast and Northwest Key cited problems: Lack of sufficient reliability testing; Questions about suitability given special local conditions (TVA, hydroelectric power...). 11

12 U.S. National Power Grid 12

13 Our ACE Wholesale Power Market Model (Java/RepastJ Framework) Modeling Goal: Operational Validity Model architecture based on operation/training manuals for market designs implemented for New England (March 2003) and Midwest (April 2005) New England/Midwest market designs meet FERC s s core WPMP architectural requirements - Independent System Operator (ISO) - Day-ahead and real-time markets for power - Congestion managed via Locational Marginal Pricing (LMP) - Financial transmission rights - ISO market oversight and market power mitigation 13

14 Basic Project Approach: Iterative Participatory Modeling See, e.g., Barreteau et al. (JASSS 2003) Stakeholders and researchers from multiple disciplines join together in a repeated looping through four stages of analysis: Field work and data collection; Role-playing games; Agent-based model development (Java/RepastJ); Intensive computational experiments. 14

15 ACE Wholesale Power Market Model (Captures salient features of FERC s s proposed design) Traders Sellers and buyers Follow market rules Learning abilities Multi-settlement process Day-ahead market (double auction, financial contracts) Real-time market (settlement of differences) Supply re-offer period AC transmission grid Independent System Operator System reliability assessments Day-ahead bid-based unit commitment Real-time settlement Sellers/buyers located at various transmission nodes Congestion managed via Locational Marginal Prices (LMP) 15

16 Example: 5-Bus Transmission Grid E Generator Transmission Line D Load Bus A C B 16

17 ACE WPM Model: Agent Hierarchy World Traders Transmission Grid Markets ISO Buyers Sellers Reliability Commitment Dispatch Settlement Load Serving Entities Generators Bilateral FTR Day- Ahead Supply Re-offers Real- Time 17

18 A Computational Seller (Generator) Public Access: // Public Methods getmarketprotocols(posting, trade, settlement); getmarketprotocols(iso market power mitigation); Methods for receiving data; Methods for retrieving Seller data. Private Access: // Private Methods Method for calculating my expected profits; Method for calculating my actual profit outcomes; Method for updating my supply offers (LEARNING). // Private Data My capacity, grid location, cost fct., current wealth ; Data recorded about external world (dispatch schedule ); Address book (communication links). 18

19 A Computational Buyer (LSE) Public Access: // Public Methods getmarketprotocols(posting, trade, settlement); getmarketprotocols(iso market power mitigation); Methods for receiving data; Methods for retrieving Buyer data. Private Access: // Private Methods Method for calculating my expected profits; Method for calculating my actual profit outcomes; Method for updating my demand bids (LEARNING). // Private Data My downstream demand, grid location, current wealth ; Data recorded about external world (dispatch schedule ); Address book (communication links). 19

20 Market Event Schedule: Simple View Start Initialization Monthly Loop FTR FTR Market Market Daily Loop Hourly Loop Real Real Time Time Market Market H=12 H=18 D+1 Day Ahead Market D+1 Supply Re-Offers End 20

21 Learners Buyers (LSEs) Submit Bids, Accept Results (0:00-23:00) (16:00) Learners Sellers (Generators) Submit Offers, Accept Results (0:00-23:00) (0:00-12:00) (0:00-23:00) (0:00-12:00) (16:00-18:00) Monthly FTR Daily (0:00-23:00) (0:00-12:00) D+1 Day-Ahead Real-Time Market (Day D) (16:00-18:00) D+1 Supply Re-Offers Accept, Process, Report, Record ISO Market Event Schedule: Detailed View 21

22 Experimental Design: Treatment Factor Ranges Cournot supply behavior Learned strategic supply beh. (Typical econ. lit. assumption) (Actual MISO/ISO-NE situation) Passive inelastic demand Learned strategic demand beh. (Typical econ. lit. assumption) (Actual MISO/ISO-NE situation) No transmission constraints Active transmission constraints (Typical econ. lit. assumption) (Actual MISO/ISO-NE situation) No financial trans. rights Fin. Trans. Rights (FTR) market (Typical econ. lit. assumption) (Actual MISO/ISO-NE situation) 22

23 Handling Learning (JReLM Module, Charles J. Gieseler, M.S. Thesis, 2005) 23

24 Example: Roth-Erev Algorithm 1. Initialize action propensities to an initial propensity value. 2. Generate choice probabilities for all actions using current propensities. 3. Choose an action according to the current choice probability distribution. 4. Update propensities for all actions using the reward for the last chosen action. 5. Repeat from step 2. 24

25 Roth-Erev Algorithm Structure Action Choice 1 Choice Propensity 1 Choice Probability 1 Action Choice 2 Action Choice 3 Choice Propensity 2 Choice Propensity 3 Choice Probability 2 Choice Probability 3 Action choice leads to a reward, followed by updating of action choice propensities based on this reward, followed by transformation of propensities into action choice probabilities 25

26 Handling Transmission Constraints DC Optimal Power Flow MISO/ISO-NE rent proprietary LMP/Dispatch solution software from commercial companies Bids/Asks DC optimal power flow LMP/Dispatch DC Optimal Power flow = Maximize total net surplus subject to mixed equality/inequality constraints reflecting branch thermal limits, node balance conditions, and generator production limits MISO/ISO-NE cannot/will not release this software 26

27 QuadProgJ Module (Sun & Tesfatsion, ISU Econ WP 06014, 2006) QuadProgJ is a general Java QP solver that can be used to solve DC Optimal Power Flow problems. Appears to be the first non-commercial open- source Java module for DC Optimal Power Flow solutions. Has handled with high accuracy and stability all small to medium-sized test cases attempted to date. Permits completion of our ACE dynamic wholesale power market with learning traders. 27

28 QuadProgJ for Solving Strictly Convex Quadratic Programming (SCQP) Problems A SCQP Problem T T Min f ( x) = x Gx + a x x T st.. C x= b C x b eq 1 2 eq T ; iq iq Inputs ( GaC,, eq, beq, Ciq, biq ) QuadProgJ Solutions * * * * ( x, λeq, λiq ; f ) 28

29 DC-OPF: 5-Node Test Case G5 Node 5 Node 4 LSE 3 G4 Node 1 Node 2 Node 3 G1 G2 LSE 1 G3 LSE 2 29

30 DC-OPF: 5-Node Test Case 24 Hour Load Distribution for 5-Node Case Load (in MWs) Hour LSE1 at Node2 LSE2 at Node3 LSE3 at Node4 30

31 DC OPF Solution: 5-Node Test Case (Branch 1-2 Congestion, Highest in Peak Hour 18) 31

32 Concluding Remarks ACE modeling (subset of constructive math) is a potentially useful tool for market design as a complement for human-subject experiments, field studies, econometrics, and classical math. ACE could facilitate iterative participatory modeling (IPM) for empirical-based modeling. But academic (open source) research via IPM faces practical challenges due to proprietary restrictions on software and data release. 32

33 Related On-Line Resources 33

34 Related On-Line Resources Cont. 34