miso market performance: an based computational test bed€¦ · miso market performance: an...

24
MISO Market Performance: An AgentBased Computational TestBed Funded in part by the ISU Electric Power Research Center Project Start Date: August 2006 Project Participants: Leigh Tesfatsion, PI (Prof. of Econ & Math, ISU) Herman Quirmbach, Co-PI (Assoc. Prof. of Econ, ISU) Hongyan Li, RA (PhD. Candidate, ECpE, ISU) http://www.econ.iastate.edu/tesfatsi/MISOEnergyGroup.htm http://www.econ.iastate.edu/tesfatsi/AMESMarketHome.htm 1

Upload: others

Post on 17-Jul-2020

5 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: MISO Market Performance: An Based Computational Test Bed€¦ · MISO Market Performance: An Agent‐Based Computational Test‐Bed Funded in part by the ISU Electric Power Research

MISO Market Performance:An Agent‐Based Computational Test‐Bed

Funded in part by the ISU Electric Power Research CenterProject Start Date: August 2006

Project Participants: Leigh Tesfatsion, PI (Prof. of Econ & Math, ISU)

Herman Quirmbach, Co-PI (Assoc. Prof. of Econ, ISU) Hongyan Li, RA (PhD. Candidate, ECpE, ISU)

http://www.econ.iastate.edu/tesfatsi/MISOEnergyGroup.htm http://www.econ.iastate.edu/tesfatsi/AMESMarketHome.htm

1

Page 2: MISO Market Performance: An Based Computational Test Bed€¦ · MISO Market Performance: An Agent‐Based Computational Test‐Bed Funded in part by the ISU Electric Power Research

Presentation OutlineProject Motivation

An Agent-Based Test Bed for Restructured Power MarketsAMES (Agent-Based Modeling of Electricity Systems)

Illustrative Experimental ResultsLMP Response to Demand-Bid Price SensitivityLMP Response to a Price-Cap on GenCo Supply Offers

Research in Progress

2

Page 3: MISO Market Performance: An Based Computational Test Bed€¦ · MISO Market Performance: An Agent‐Based Computational Test‐Bed Funded in part by the ISU Electric Power Research

Project MotivationIn April 2003, FERC proposed a new market design for U.S. wholesale power markets.

About 50% of U.S. electric power generating capacity now operates under a variant of the FERC market design.

These restructured wholesale power markets are extremely complicated, involving

Physical constraintsInstitutional arrangementsBehavioral dispositions of human participants

Difficult to model and study these markets using standard analytical and statistical tools.

3

Page 4: MISO Market Performance: An Based Computational Test Bed€¦ · MISO Market Performance: An Agent‐Based Computational Test‐Bed Funded in part by the ISU Electric Power Research

The AMES Test BedAMES is a computational wholesale power market populated by interacting software “agents.”

Agents range from passive world features (e.g., transmission grids) to sophisticated decision makers with communication and learning capabilities (e.g., market operators and traders).

AMES permits the systematic experimental testing of the FERC market design (e.g., as implemented by MISO)new/modified market design features

AMES = Agent-based Modeling of Electricity SystemsSoftware downloads, manuals, and publications are available athttp://www.econ.iastate.edu/tesfatsi/AMESMarketHome.htm

4

Page 5: MISO Market Performance: An Based Computational Test Bed€¦ · MISO Market Performance: An Agent‐Based Computational Test‐Bed Funded in part by the ISU Electric Power Research

AMES Target Features and Release History

Research/teaching/training test bed (2-500 pricing nodes)

Operational validity (structure, rules, behavioral dispositions)

Permits dynamic testing with learning traders

Permits intensive experimentation with alternative scenarios

Open-source Java implementation (full access to code)

Flexible and modular (easily modified test bed features)

V1.31 released (IEEE PES General Meeting, 2007)

V2.0 in final testing stage (to be released at IEEE PES Gen Meet, 2008)

5

Page 6: MISO Market Performance: An Based Computational Test Bed€¦ · MISO Market Performance: An Agent‐Based Computational Test‐Bed Funded in part by the ISU Electric Power Research

AMES Flexible Modular Architecture

Market protocols and AC transmission grid structureModularized class structure (entirely in Java) permits easy experimentation with alternative market rules and grid constraints.

Learning representations for tradersJava Reinforcement Learning Module (JReLM)“Tool box” permitting experimentation with a wide variety of learning methods (reactive reinforcement learning, anticipatory learning,…)

Optimal power flow formulationsJava DC Optimal Power Flow Module (DCOPFJ)Java AC Optimal Power Flow Module (ACOPFJ in final testing stage)

Graphical user interfaceEasy parameter editing Customizable chart/table input and output displays5-bus/30-bus test cases

6

Page 7: MISO Market Performance: An Based Computational Test Bed€¦ · MISO Market Performance: An Agent‐Based Computational Test‐Bed Funded in part by the ISU Electric Power Research

AMES Architecture: Current Implementation(based on business practices manuals for MISO/ISO-NE)

TradersGeneration Companies (sellers)Load-Serving Entities (buyers)Learning abilities

Two-Settlement ProcessDay-ahead market (double auction, financial contracts)Real-time market (settlement of differences)

AC Transmission GridGeneration Companies (GenCos) & Load-Serving Entities (LSEs) located at various transmission nodesCongestion managed via Locational Marginal Pricing (LMP) determined by ISO via DC or AC optimal power flow (OPF)

7

Independent System Operator System reliability assessmentsDay-ahead bid/offer-based unit commitmentReal-time dispatch

Page 8: MISO Market Performance: An Based Computational Test Bed€¦ · MISO Market Performance: An Agent‐Based Computational Test‐Bed Funded in part by the ISU Electric Power Research

Activities of AMES ISO During Each Operating Day D:Timing Adopted from Midwest ISO (MISO)Timing Adopted from Midwest ISO (MISO)

00:00

11:00

18:00

23:00

Real-time

(spot)market

forday D

Day-ahead marketfor day D+1

ISO collects bids/offers from LSEs and GenCos

ISO evaluates demand bids and supply offers

ISO solves D+1 DC OPF and posts D+1 commitment

and LMP schedule

Day-ahead settlementReal-time settlement

Page 9: MISO Market Performance: An Based Computational Test Bed€¦ · MISO Market Performance: An Agent‐Based Computational Test‐Bed Funded in part by the ISU Electric Power Research

AMES Graphical User Interface (GUI):Tool Bar and Menus

9

Page 10: MISO Market Performance: An Based Computational Test Bed€¦ · MISO Market Performance: An Agent‐Based Computational Test‐Bed Funded in part by the ISU Electric Power Research

AMES GUI: Output Chart Display Illustration

10

Page 11: MISO Market Performance: An Based Computational Test Bed€¦ · MISO Market Performance: An Agent‐Based Computational Test‐Bed Funded in part by the ISU Electric Power Research

Illustrative Experimental Findings (AMES Version 2.0)

Hongyan Li, Junjie Sun, and Leigh Tesfatsion, “Dynamic LMP Response Under Alternative Price-Cap and Price-Sensitive Demand Scenarios”

Proceedings, IEEE Power Engineering Society General Meeting, Pittsburgh, PA, July 2008, to appear

www.econ.iastate.edu/tesfatsi/DynamicLMPResponse.IEEEPES2008.LST.pdf

11

Page 12: MISO Market Performance: An Based Computational Test Bed€¦ · MISO Market Performance: An Agent‐Based Computational Test‐Bed Funded in part by the ISU Electric Power Research

LMP Response to Price-Sensitive Demand and to a Price Cap on GenCo Supply Offers

LSE demand bids for day-ahead market are mixtures of fixed (price-insensitive) demands and price-sensitive demands.

Currently in MISO, price-sensitive demand is only about 1% of the total bid-in demand for the day-ahead market.

What is LMP response if we systematically change the percentage of price-sensitive demand?

What if the ISO imposes a price cap on GenCo supply offersfor the day-ahead market?

Do LMPs have a controllable upper limit?Will this give profit-seeking GenCos an incentive to report smaller-than-true max capacities?

12

Page 13: MISO Market Performance: An Based Computational Test Bed€¦ · MISO Market Performance: An Agent‐Based Computational Test‐Bed Funded in part by the ISU Electric Power Research

Form of GenCo Supply Offers

Hourly supply offer for each GenCo i = Reported linear marginal cost function over a reportedoperating capacity interval for real power pGi (in MWs):

MCRi(pGi) = aR

i + 2bRi pGi

CapiL ≤ pGi ≤ Capi

RU

GenCos can report higher-than-true marginal costs and/or withhold capacity.

13

Page 14: MISO Market Performance: An Based Computational Test Bed€¦ · MISO Market Performance: An Agent‐Based Computational Test‐Bed Funded in part by the ISU Electric Power Research

Form of LSE Demand Bids

Hourly demand bid for each LSE j = (Fixed demand bid, Price-sensitive demand bid)

Fixed demand bid = pFLj (in MWs)

Price-sensitive demand bid= Linear demand function for real power pS

Lj (in MWs) overa purchase capacity interval:

Dj(pSLj) = cj - 2dj pS

Lj

0 ≤ pSLj ≤ SLMaxj

14

Page 15: MISO Market Performance: An Based Computational Test Bed€¦ · MISO Market Performance: An Agent‐Based Computational Test‐Bed Funded in part by the ISU Electric Power Research

Construction of R Ratio for Measuring Percentage of Price-Sensitive Demand

15

Page 16: MISO Market Performance: An Based Computational Test Bed€¦ · MISO Market Performance: An Agent‐Based Computational Test‐Bed Funded in part by the ISU Electric Power Research

5‐Bus Test Grid

Extension of static test case (by John Lally) used in ISO-NE/PJM training materials

Page 17: MISO Market Performance: An Based Computational Test Bed€¦ · MISO Market Performance: An Agent‐Based Computational Test‐Bed Funded in part by the ISU Electric Power Research

Average Effects of R Changes for Benchmark Case[No Supply-offer Price Cap and No GenCo Learning]

17

Lerner Index

Page 18: MISO Market Performance: An Based Computational Test Bed€¦ · MISO Market Performance: An Agent‐Based Computational Test‐Bed Funded in part by the ISU Electric Power Research

Average Effects of R Changes with GenCo Learning[No Supply-offer Price Cap]

18

GenCos use stochastic reactive reinforcement learning to select their supply offers for the day-ahead market.

Page 19: MISO Market Performance: An Based Computational Test Bed€¦ · MISO Market Performance: An Agent‐Based Computational Test‐Bed Funded in part by the ISU Electric Power Research

Summary of Price-Sensitive Demand Experiments

BOTTOM LINE:Even with 100% price-sensitive demand bids (R=1.0),average LMP is much higher under GenCo learning !

NEEDED:Active demand-side bidding from LSEs reflecting better

integration of wholesale and retail marketsCountervailing power (active supply offers AND

active demand bids at wholesale level) shouldresult in more competitive pricing.

19

Page 20: MISO Market Performance: An Based Computational Test Bed€¦ · MISO Market Performance: An Agent‐Based Computational Test‐Bed Funded in part by the ISU Electric Power Research

LMP Response to a Supply-Offer Price Cap [100% Fixed Load (R=0.0)]

20

With GenCo learning, average LMP successively declines as a stricter supply-offer price cap is imposed.

Page 21: MISO Market Performance: An Based Computational Test Bed€¦ · MISO Market Performance: An Agent‐Based Computational Test‐Bed Funded in part by the ISU Electric Power Research

LMP Response to a Supply-Offer Price Cap[with 100% Fixed Load (R=0.0) and with GenCo Learning]

21

Page 22: MISO Market Performance: An Based Computational Test Bed€¦ · MISO Market Performance: An Agent‐Based Computational Test‐Bed Funded in part by the ISU Electric Power Research

Average LMP Volatility and Spiking[with 100% Fixed Load (R=0.0) and with GenCo Learning]

22

Page 23: MISO Market Performance: An Based Computational Test Bed€¦ · MISO Market Performance: An Agent‐Based Computational Test‐Bed Funded in part by the ISU Electric Power Research

Summary of Supply-Offer Price Cap Experiments

BOTTOM LINE:Supply-offer price caps can lead to increased LMP volatility/spiking and inadequacy events (S<D) around peak demand hours even though average LMP declines!

23

Page 24: MISO Market Performance: An Based Computational Test Bed€¦ · MISO Market Performance: An Agent‐Based Computational Test‐Bed Funded in part by the ISU Electric Power Research

Research in ProgressFurther development of AMES test bed

More complete modeling of Security Constrained Unit Commitment (SCUC) to include: start-up costs, ramping constraints, minimum run times and down times, and security constraints.Fuller implementation of two settlement system (real-time and day-ahead markets) to handle uncertainty in load, generation, and transmission network operating conditions.Additional testing of new AC OPF module. Extension of learning toolkit (JReLM) to include additional learning methods, with LSEs and GenCos both able to learn.

Use of AMES to explore MISO market performanceTests of two settlement system under normal and insecure conditions.Comparing AC OPF solutions with DC OPF approximate solutions.Dynamic market performance when all traders have learning capabilities.

24