Argonne National LaboratoryDecision and Information Sciences Division
Center for Complex Adaptive Agent Systems Simulation
Modeling the Restructured Illinois Electricity Modeling the Restructured Illinois Electricity Market as a Complex Adaptive SystemMarket as a Complex Adaptive System
Charles M. MacalCharles M. MacalGale A. BoydGale A. Boyd
Richard R. Richard R. CirilloCirilloGuenterGuenter ConzelmannConzelmann
Michael J. NorthMichael J. NorthPrakashPrakash R. R. ThimmapuramThimmapuram
Thomas D. Thomas D. VeselkaVeselkaCenter for Complex Adaptive Agent System Simulation (CASCenter for Complex Adaptive Agent System Simulation (CAS22))
Decision & Information Sciences DivisionDecision & Information Sciences DivisionArgonne National LaboratoryArgonne National Laboratory
[email protected]@anl.gov, , www.cas.anl.govwww.cas.anl.gov
Presented at thePresented at the2424ndnd Annual North American Conference of the USAEE/IAEEAnnual North American Conference of the USAEE/IAEE
Energy, Environment and Economics in a New EraEnergy, Environment and Economics in a New EraJuly 8July 8--10, 200410, 2004
Washington, D.C.Washington, D.C.
Argonne National LaboratoryDecision and Information Sciences Division
Center for Complex Adaptive Agent Systems Simulation
Outline
Deregulated electric power markets and the agent modeling approachEMCAS modelInvestigations into market powerLessons learned
Argonne National LaboratoryDecision and Information Sciences Division
Center for Complex Adaptive Agent Systems Simulation
Electricity Markets Are Changing
The new deregulated electricity markets are founded on decentralized competitive decision-making
– Markets have greater complexity– Markets have more players– Participants have different (often
conflicting) objectives and decisioncharacteristics or risk preferences
Conventional modeling approachesmust be extended to account forthe new reality of decentralizedcompetitive decision-making
These extensions must beconsistent with existingapproaches and often include them in the larger modeling framework
LoadAggregators
LoadAggregators
CustomersCustomersGeneration Companies
BrokersBrokersBrokersBrokers
LDC
LDC
LDC
LDC
LDCLDC
LDCLDC
PX
ISO
PX
ISO
PX
ISO
PX
ISO
bids
bids
bids
bids
contracts
contracts
contracts
Argonne National LaboratoryDecision and Information Sciences Division
Center for Complex Adaptive Agent Systems Simulation
Why Agents?
The rules of business and social interaction are at least as important as the rules of physics when it comes to the generation, sale, and delivery of electrical power
Agents operating within an agent framework can be used to model decentralized competitive decision-making
Agent frameworks allow groups of agents to interact in complex, dynamic waysLearning and adaptation of agent behavior can be modeledTransient conditions of the system can be studied in addition othe long-run (equilibrium?) conditionsAlternative market rules can be tested
Argonne National LaboratoryDecision and Information Sciences Division
Center for Complex Adaptive Agent Systems Simulation
Agents Are Autonomous, Self-directed, Decision-making Units
• Agents are heterogeneous• Goals• Behaviors• Attributes• Available accumulated resources
• Decision rules vary by agent• Sophistication of rules• Cognitive “load”• Internal models of the external world• Memory employed
• What is the effect of individual agent behavior (rules) on the system?• Do certain types of agents dominate?• Does the system evolve toward a stable
mix of agent types?
Agent• Attributes• Rules of behavior• Memory• Sophistication• Resources
Argonne National LaboratoryDecision and Information Sciences Division
Center for Complex Adaptive Agent Systems Simulation
Physical agents– Generators– Transmission buses– Transmission lines
Decision-making agents– Consumers– Demand agents– Distribution companies– Generation companies– Transmission companies– ISO/RTOs– Regulators
The EMCAS Model Uses A Variety of Agents to ModelDecentralized Electricity Markets
Argonne National LaboratoryDecision and Information Sciences Division
Center for Complex Adaptive Agent Systems Simulation
Generation Company Agents Consider Many Factors in Proposing Bids for the Day-Ahead Market
Argonne National LaboratoryDecision and Information Sciences Division
Center for Complex Adaptive Agent Systems Simulation
EMCAS Operates atMultiple Time Scales and Decision Cycles
Planning: –Day-Ahead Planning:
Agents determine market allocations for selling productsBilateral contracts are formed with individual demand agents, and energy bids are sent into the ISOAgents make unit commitment schedules for the next day
Dispatch: Hourly/Real-Time Dispatch:–Power plants are operated as directed by the ISO in accordance with prior market arrangements made under bilateral contracts and in energy and ancillary service markets
Weekly and Monthly Adjustments–Bilateral contracts are made with individual demand agents and are sent to the ISO for approval
–Marketing strategies are adjusted–Adjustments can be made to unit maintenance schedules
Argonne National LaboratoryDecision and Information Sciences Division
Center for Complex Adaptive Agent Systems Simulation
DECISION OUTPUT
• Bid accepted/rejected• Unit utilization• Unit profitability• Market price vs. bid price• Weather & Load
EMCAS Agents Make Decisions Based on BothPast Experiences and Future Expectations
• Own unit availability• Prices• Weather• Load
• Competing unit availability• Own cost structure• Market rules • Bid structure: capacity
blocks for different markets
• Bid prices for each block and market
LOOK BACK (Short andLong-Term Memory)
TIME
AgentAgentDecisionDecision
RulesRules
LOOK SIDEWAYS
LOOK AHEADEXAMPLE: GENERATIONCOMPANY AGENT
Argonne National LaboratoryDecision and Information Sciences Division
Center for Complex Adaptive Agent Systems Simulation
Projection function– Forecasts next day weather, system demand,
and system available generation capacity– Makes information available to all agents
Pool market function– Operates the pool market for energy
and ancillary services– Administers price setting rules such as LMP
pricing or pay-as-bid pricingScheduling function
– Accepts or rejects pool market bids andbilateral contracts using conventionalload flow and optimization tools
Dispatching function– Dispatches the generators in real time to match the demand– Maintains the necessary security requirements
Settlement function– Applies settlement rules selected by the user to calculate the payments to and receipts from
the generating companies, demand agents, and transmission companies
The EMCAS ISO/RTO Agent PerformsSeveral Functions
Argonne National LaboratoryDecision and Information Sciences Division
Center for Complex Adaptive Agent Systems Simulation
The Physical System is Represented at the Bus/Branch Level While a Network Reduction Allows Zonal Market Simulation
Working with the University of Illinois and the PowerWorld AC power flow model, an equivalenced DC OPF (optimal power flow) model is derived and embedded in EMCAS– In-state equivalenced network solved for 24-hour market– Includes equivalenced out-of-state representations
Major topology changes (e.g., substantial line outages) requirenetwork reduction to be rerun
917
896
896
542
922
891
895
811
895
891893
200
200
2663
542
1109
2003441
468
1604
2146
1973
2622
2685
2489
815
2269
2718
112
1409
1348
1289
1193
569
200
2551
717
3010
200
531
1109
200
1008
9977
6377
1751
2409
200
578
468
1973
2394
2971
2975
2409
569
717
3081
3694
2717
200
2509
538
920
1728
200
535
504
1604
200
9202409
186
1861751
27958
543
1664
2006576
200
1621
786
968
9681497
1604
14742
11414
18667
195
195
2152
2146
2409
2188
3105
2409
836
434
AECI
AEP
ALTE
ALTW
AMRNA
AMRNB
AMRNC
AMRND
CILC
CIN
CWLP
DPC
EEI
ENLC
IPA
IPB
IPC
IPD
MEC
MIPUNIA
NIB
NIC
NID
NIE
NIF
NIG
NIPS
SIPC
TVA
WEC
Network Reduction fromBus-Level to Zonal Level
Argonne National LaboratoryDecision and Information Sciences Division
Center for Complex Adaptive Agent Systems Simulation
Application: Electric Power MarketsHow will Illinois fare under electricity deregulation: Jan. 1, 2007?
1. Will power transmission capacity be adequate, or is congestion likely? Under what conditions?
2. Will transmissions constraints on the power grid create regional imbalances in supply and demand?
3. Will imbalances create pockets of market power and potentially drive up locational electricity prices?
AMRN - A
AMRN- B
AMRN- C
AMRN - D
AMRN- E
CILC
CWLP
EEI
IP- A
IP- B
IP- C
IP- D
NI- A NI- B
NI- C
NI- D
NI- E
NI- F
NI- G
SIPC
Buses : 1856Bus locations : 441Links displayed : 1283
Transmission Line Capacity
0 £ kV < 125
125 £ kV < 150
150 £ kV < 300
300 £ kV < 400
400 £ kV
Argonne National LaboratoryDecision and Information Sciences Division
Center for Complex Adaptive Agent Systems Simulation
How Would We Go About Using Such A Model to Answer the Question of Whether There Is Market Power?
Establish a base case that excludes strategic behavior – bids based on production costEstablish reasonable assumptions on scenario variables– Weather and effect on peak load– Unit outage assumptions– New capacity additions and transmission network adjustments– Agree on definitions (short-run, long-run), e.g., production cost – Identify credible strategies that agents might employ
Test the effect of these strategies incrementally with the modeland compare with the base case
Argonne National LaboratoryDecision and Information Sciences Division
Center for Complex Adaptive Agent Systems Simulation
Production-Cost Case Analysis
The following assumptions and market rules are assumed for the analysis year:Locational marginal prices (LMP) are paid to GenCos;DemCos pay the load-weighted average zonal price;There is a day-ahead market (DAM) administered by the ISO/RTO for both energy and ancillary services where GenCos and DemCos can participate;Bids are unregulated;Bilateral contracts between suppliers (GenCos) and purchasers (DemCos) are not allowed; all power is purchased in the spot markets;Generation companies sell electricity based entirely on the production costs of the generators, including fixed costs;Demand companies serve firm uninterruptible load; unserved energy is due only to forced outage conditions and not to any market considerations; consumers are assumed to have no response to electricity prices; andTransmission companies do not employ any strategic business behavior; their revenue comes in two forms: a transmission use charge of $3 per MWh and a transmission congestion payment.
Argonne National LaboratoryDecision and Information Sciences Division
Center for Complex Adaptive Agent Systems Simulation
Strategies for GenCo Agents
GenCo agents have various levers to base their strategies on– Capacity to offer (bid) into the market– Price to offer the capacity
Witholding– Physical Witholding– Economic Witholding
Price Probing: probe the market for weaknesses or flaws– Discover if you are the marginal supplier– Discover who is the marginal supplier
A Host of Strategies:– Dynamic incremental pricing– Fixed incremental pricing– Bid production cost– Bid low to ensure dispatch (EDF, for spot market only)– Bid high to increase the market clearing price– Bid last increment of capacity at high price– Adjust prices based on past performance (bid price based on moving average price)
Argonne National LaboratoryDecision and Information Sciences Division
Center for Complex Adaptive Agent Systems Simulation
Testing Strategies for GenCo Agents: Witholding
Physical Witholding: Capacity taken off line in order to improve business position– Witholding capacity during periods of low prices, in which costs cannot be
recovered– Witholding capacity for certain units increases LMPs for the system and the
profits of other units in a companies generator portfolioEconomic Witholding: Generation capacity not taken off line –make available to the market at increased pricesFators to Consider in Implementing a Witholding Strategy– Unit capacity: larger units tend to affect market prices more if witheld– Unit location in transmission network: units in areas of transmission
congestion have larger impacts on the market if the transmission system cannot allow replacement capacity to be utilized
– Availability of replacement capacity: Availability of replacement capacity and its price, will determine how the market will respond to physical witholding
Argonne National LaboratoryDecision and Information Sciences Division
Center for Complex Adaptive Agent Systems Simulation
Physical Witholding Experiments
Experiment: GenCos physically withold units, observe affect on LMPswhich results in changes in revenues. Which units to withold?
Witholding a single unit of a company on peak load days, – Preliminary results suggest that this strategy will not in general increase a company’s
profits– Loss of revenue from the unit witheld was not offset by higher prices paid to the
company’s units still operatingWitholding multiple company units, based on unit profitability (smallest profit potential)
– Preliminary results suggest that this strategy will not in general increase a company’s profits
Witholding multiple company units, using system reserve and unit location as criteria
– Withholding done only when the system reserve was expected to be at a low point– Witholding units located at critical transmission network points (those with high LMPs)– Preliminary results suggest that this strategy can increase company operating profit
and consumer costs considerably
Argonne National LaboratoryDecision and Information Sciences Division
Center for Complex Adaptive Agent Systems Simulation
Economic Witholding Experiments
Experiment: GenCos raise bid prices above production costs, observe affect on LMPs which results in changes in revenues. Which units to economically withold and for which hours?Company increases prices for a single unit of a company on peak load days
– Preliminary results suggest that this strategy will not in general increase a company’s profits
– Adequate generation and transmission capacity to bring cheaper units on line
– A few units critical during peak hours provide small increases in operating profit
Company increases prices for all company units– Preliminary results suggest that this strategy will not in general increase a
company’s profits if done for all units and all hours of a peak load day– May increase profits for a few larger companies if done for peak hours only
Argonne National LaboratoryDecision and Information Sciences Division
Center for Complex Adaptive Agent Systems Simulation
Testing Strategies for GenCo Agents: Price Probing
Price Probing: Test the bounds of the prices that can be offered in the market before competitors are able to capture market shareRaising bid prices can result in increased profits if there is:– Insufficient alternative generation capacity, and/or– Inadequate transmission capacityFixed Increment Probing– A company raises prices by the same amount every day until its units are
not selected and then retreats on pricesDynamic Increment Probing– Prices on units are raised by varying amounts depending on
1. evaluation of a company’s operating profit2. the projected system status, and3. whether specific units in their portfolio are located at critical points in the network
Argonne National LaboratoryDecision and Information Sciences Division
Center for Complex Adaptive Agent Systems Simulation
Potential Future Load Pockets
Base Case• Preliminary results suggest that higher LMPs are experienced through
some portions of the study area• Cheaper generation available in some areas does not help relieve the
higher prices in other parts of the study area due to transmission limitations
Argonne National LaboratoryDecision and Information Sciences Division
Center for Complex Adaptive Agent Systems Simulation
Base Case
0
10
20
30
40
50
60
70
80
90
Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec
LMP
[$/M
Wh]
MaxLMPs
MinLMPs
High LMPs due toForced Outages
High LMPs due to high load
Spread of LMPs due totransmission congestion
Projected Monthly Minimum and Maximum Hourly LMPs for All Zones
• Congestion-related dispatch of units out of bid merit order leads to LMP differences across the system (note, no strategic bidding)
Argonne National LaboratoryDecision and Information Sciences Division
Center for Complex Adaptive Agent Systems Simulation
Identifying Market Power
0
50
100
150
200
250
1 2 3 4 5 6 7 8
Peak
-Day
24-
Hou
r Gen
erat
ion
[GW
h]
0.0
2.0
4.0
6.0
8.0
10.0
Peak
-Day
24-
Hou
r Ope
ratin
g Pr
ofit
[$ m
illio
n]Generation (2pm-6pm Strategy)Generation (All-Day Strategy)Profits (2pm-6pm Strategy)Profits (All-Day Strategy)
BaseCase
25%Above Cost
50%AboveCost
100%AboveCost
150%AboveCost
400%AboveCost
650%AboveCost
200%AboveCost
GenCo with No Market Power
-20
0
20
40
60
80
100
Peak
-Day
24-
Hou
r Gen
erat
ion
[GW
h]-0.2
0.0
0.2
0.4
0.6
0.8
1.0
Peak
-Day
24-
Hou
r Ope
ratin
g Pr
ofit
[$ m
illio
n]Generation (2pm-6pm Strategy)Generation (All-Day Strategy)Profits (2pm-6pm Strategy)Profits (All-Day Strategy)
BaseCase
25%Above Cost
50%AboveCost
100%AboveCost
400%AboveCost
200%AboveCost
GenCo with Market Power
• GenCo bid prices are increased to various levels above production cost– All-day– Afternoon only
Argonne National LaboratoryDecision and Information Sciences Division
Center for Complex Adaptive Agent Systems Simulation
Lessons Learned from Agent-based Modeling of Electric Power Market
Agent-based modeling can be a useful adjunct to understanding the deregulated electric power market– Important to have the agent model produce a base case that is comparable
to what traditional modeling approaches would produceModeling (and validating) the interface/interplay between the physical system and the economic system is essential for credibilityAgent strategies that are based on bounded rationality in lieu of a theory of agent behavior require special modeling approaches:– Agent strategy hypothesizing, simplicity, screening, and support– Sensitivity analysis and parameter space exploration (lots of simulation
runs)– Issue of non-attribution of agent behaviors to particular agents in the system
(an agent could behave that way, but would they?)
Argonne National LaboratoryDecision and Information Sciences Division
Center for Complex Adaptive Agent Systems Simulation
Modeling the Restructured Illinois Electricity Market Modeling the Restructured Illinois Electricity Market as a Complex Adaptive Systemas a Complex Adaptive System
Questions?
Charles M. MacalArgonne National Laboratory
Center for Complex Adaptive Agent Systems Simulation (CAS2)[email protected]
www.cas.anl.gov