A g e n t s
a n d
M a r k e t s
MASCEM: A Multiagent
System That Simulates
Competitive Electricity
Markets
Isabel Praça, Carlos Ramos, and Zita Vale, Polytechnic Institute of Porto
Manuel Cordeiro, University of Trás-os-Montes e Alto Douro
A
round the world, the electricity industry, which has long been dominated by
vertically integrated utilities, is experiencing major changes in the structure
of its markets and regulations. Owing to new regulations, it’s evolving into a distributed
industry in which market forces drive electricity’s price. The industry is becoming
MASCEM, a multiagent
simulator system,
can be a valuable
framework for
evaluating new rules,
new behavior, and new
participants in the
numerous electricity
markets that are
moving toward
liberalization and
competition.
54
competitive; a market environment is replacing the
traditional centralized-operation approach. This
transformation is often called the deregulation of the
electricity market.
Many electricity companies have split into several
companies that specialize in energy generation,
transmission, or distribution. These companies need
to adopt new management approaches to survive in
this market. The industry needs new software
tools—such as price-based unit commitment and
price-bidding tools—to support new market activities. Also, it needs new modeling approaches that
simulate how electric power markets might evolve
over time and how market participants might react
to the changing economic, financial, and regulatory
environment in which they operate.
To study electricity market behavior and evolution, we developed MASCEM,1 a multiagent simulator system where agents represent market entities
such as generators, consumers, and market operators, and new participants such as traders. Agents can
establish their own objectives and decision rules.
Moreover, they can adapt their strategies as the simulation progresses on the basis of previous efforts’
success or failure. The simulator probes the effects
of market rules and conditions by simulating the participants’ strategic behavior.
As a decision support tool, our simulator includes
several types of negotiation mechanisms found in
electricity markets to let the user test them and learn
the best way to negotiate in each one. The simulator
is flexible; the user completely defines the model he
or she wants to simulate, including the number of
agents, each agent’s type and strategies, and the market type. (See the “Related Work” sidebar for other
approaches to simulation.) Electric companies as
well as regulator entities can use MASCEM to analyze
the impact of their potential decisions.
Market structure
An electricity market’s main objectives are to
ensure the system’s secure and efficient operation and
to decrease the cost of electricity through competition. Several market structure models exist that could
help achieve this goal.2 The market environment typically consists of a pool, as well as a floor for bilateral contracts.
A pool, or power exchange, is a marketplace
where electricity-generating companies submit production bids and their corresponding market prices,
and consumer companies submit consumption bids
(see the “Pool Mechanisms” sidebar). A market
operator regulates the pool. The market operator uses
a market-clearing tool to set the market price every
hour; this results in a market-clearing price and a set
of accepted production and consumption bids. In
pools, an appropriate market-clearing tool is an auction mechanism—the most common being a standard uniform auction.
Bilateral contracts are negotiable agreements
between sellers and buyers (or traders) about power
supply and receipt. The bilateral-contract model is
1094-7167/03/$17.00 © 2003 IEEE
Published by the IEEE Computer Society
IEEE INTELLIGENT SYSTEMS
Related Work
Some other approaches to agent-based simulation applications
for competitive electricity markets are more targeted or limited
than MASCEM. PowerWeb (http://stealth.ee.cornell.edu/powerweb)
considers single uniform auctions with fixed demand. The Auction Agents for the Electric Power Industry project (www.agentbuilder.com/Documentation/EPRI) implements a Dutch auction,
which is a specific type of auction mechanism. SEPIA (Simulator for
Electric Power Industry Agents) studies just bilateral contracts
(www.htc.honeywell.com/projects/sepia).
The works of Derek Bunn and Fernardo Oliveira;1 FrançoisRégis Monclar and Richard Quatrain;2 and James Nicolaisen,
Valentin Petrov, and Leigh Tesfatsion3 are relevant to our
research in agent-based simulation; however, they discuss
only the market in England and Wales. Jorge Villar and Hugh
Rudnick present another important simulation application,4
focusing particularly on hydroelectric power station parameters. Javier Contreras and his colleagues present an interesting
lab experience that shows the practical utility of electricity
market simulators.5 The Electricity Market Complex Adaptive
System (EMCAS)6 is a relevant and interesting e-laboratory for
testing regulatory structures where agents have strategies
based on learning and adaptation.
flexible; negotiating parties can specify their
own contract terms.
The hybrid model combines features of
pools and bilateral contracts. In this model,
a pool isn’t mandatory, and customers can
either negotiate a power supply agreement
directly with suppliers or accept power at the
pool market price. In this model, participants
can negotiate in the pool and also through
bilateral contracts. This model therefore
offers customer choice.
References
1. D. Bunn and F.S. Oliveira, “Agent-Based Simulation—An Application to the New Electricity Trading Arrangements of England and
Wales,” IEEE Trans. Evolutionary Computation, vol. 5, no. 5, Oct.
2001, pp. 493–503.
2. F.R. Monclar and R. Quatrain, “Simulation of Electricity Markets: A MultiAgent Approach,” Proc. 2001 Int’l Conf. Intelligent System Application to
Power Systems (ISAP 01), 2001, pp. 207–212.
3. J. Nicolaisen, V. Petrov, and L. Tesfatsion, “Market Power and Efficiency in a Computational Electricity Market with Discriminatory
Double-Auction Pricing,” IEEE Trans. Evolutionary Computation,
vol. 5, no. 5, Oct. 2001, pp. 504–523.
4. J. Villar and H. Rudnick, “Hydrothermal Market Simulator Using
Game Theory: Assessment of Market Power,” IEEE Trans. Power
Systems, vol. 18, no. 1, Feb. 2003, pp. 91–98.
5. J. Contreras et al., “Power Engineering Lab: Electricity Market Simulator,” IEEE Trans. Power Systems, vol. 17, no. 2, May 2002, pp. 223–228.
6. M. North et al., “E-Laboratories: Agent-Based Modeling of Electricity
Markets,” Proc. Am. Power Conf., PennWell Corp., 2002; www.dis.anl.
gov/msv/msv_cas.html.
Market facilitator
Buyers
Sellers
Market operator
Trader
Pool
~
Network operator
~
Overview of MASCEM
Our market simulator is related to the
electricity day-ahead market, with 24 negotiation periods each day. It includes these
types of agents: a market facilitator, sellers,
buyers, traders, a market operator, and a network operator (see Figure 1).
The market facilitator coordinates the
simulated market and ensures that it functions correctly. It knows the identities of all
agents in the market, regulates negotiation,
and ensures the market operates according
to established rules. Before entering the market, agents must first register with the market
facilitator, specifying their role and services.
Sellers and buyers are the two key players
in the market, so we devote special attention
to them—particularly to their objectives and
strategies to reach them. Sellers represent
entities able to sell electricity in the market,
such as generating companies holding elecNOVEMBER/DECEMBER 2003
~
Bilateral
contracts
~
Figure 1. Multiple agents in MASCEM.
tricity production units. Buyers represent
electricity consumers and electricity distribution companies. The user completely
defines the number of sellers and buyers in
each scenario and must specify their intrinsic
and strategic characteristics. “Intrinsic characteristics” refers to the agent’s individual
knowledge related to limit price, preferred
prices, and available capacity (or consumption needs if it’s a buyer). “Strategic characteristics” refers to the strategies the agent will
www.computer.org/intelligent
use to reach the objective of selling the available capacity at the best price (if a seller) or
buying the needed power (if a buyer). Seller
agents will compete with each other because
they’re all interested in selling their available
capacity at the highest-possible market quote.
On the other hand, seller agents will cooperate with buyer agents to establish an agreement that’s profitable for both. This is a rich
domain for which it’s possible to develop and
test several algorithms and negotiation mech55
A g e n t s
a n d
M a r k e t s
Pool Mechanisms
In pools, the most common type of negoAsymmetric market
Symmetric market
tiation is a standard uniform auction.1,2 If
Kilowatts
Kilowatts
only suppliers can compete in the pool, it’s
called an asymmetric market. If both suppliBuyers'
Sellers'
bids
ers and buyers can compete, it’s a symmetric
bids
Demand
Sellers' bids
market (similar to a double auction in aucD
tion theory). Figure A shows both markets.
In an asymmetric market, the suppliers
present their bids, and the market operator
(the entity responsible for the pool’s nondiscriminatory functioning) organizes them
Cents per
Cents per
Market price
Market price
kilowatt-hour
kilowatt-hour
starting with the lowest price and moving
(1)
(2)
up. The consumers reveal their needs to set
up the demand. Once the market operator
knows the demand, it accepts the suppliers’
Figure A. Two types of markets: (1) asymmetric; (2) symmetric.
bids starting from the lowest and accepts as
many as are necessary to fill the demand.
ket price and every consumer offering prices higher than the
The market price—to be paid to all accepted suppliers—is that
market price will be accepted.
of the last accepted bid (the one with the highest price).
In a symmetric market, suppliers and consumers both submit
References
bids. The market operator orders the selling and demand offers:
selling bids start with the lowest price and move up, and demand
1. G. Sheblé, Computational Auction Mechanisms for Restructured
bids start with the highest price and move down. Then, the proPower Industry Operation, Kluwer Academic Publishers, 1999.
posed bids form the supply and demand step curves, and the
point at which both curves intersect determines the market
2. P. Klemperer, “Auction Theory: A Guide to the Literature,” J. Economic Surveys, vol. 13, no. 3, July 1999, pp. 227–286; www.nuff.
price, paid to all accepted supplier and consumers. The bids of
ox.ac.uk/economics/people/klemperer.htm.
every supplier offering prices lower than the established mar-
anisms for cooperation and competition.
The increased competition creates opportunities for new players or agents to enter the
market, such as traders. The introduction of
traders promotes liberalization and competition and simplifies sellers’ and buyers’ dealings with each other and with the market operator. Traders are intermediaries between
consumers and suppliers. They can act as
retailers, brokers, aggregators, or marketers.
Other simulators don’t include this agent type.
The network operator is specific to the
application domain of electricity markets. It
is responsible for the system’s security and
all involved technical constraints. One network operator is present in every simulation
and must be informed of every contract
established—either through bilateral contracts or through the pool. It analyzes the contract’s technical viability from the power system viewpoint (for example, the feasibility
of power flow to meet all needs) and manages to solve congestion situations.
The market operator manages the pool
mechanism. This agent is present only in
pool or hybrid market simulations. We’ll talk
more about this in the next section.
56
The pool
The market operator starts the negotiation by sending a request for proposals
at the beginning of each negotiation period.
All interested agents—sellers, buyers, and
traders—reply by sending bids to the pool.
Then the market operator organizes the bids
received, determines the market price, and
determines the accepted and rejected bids. It
performs bid matching according to the network operator’s technical viability analysis.
After the market operator processes the bids
and establishes the market price, it communicates the results to each pool participant.
Bilateral contracts
Bilateral contracts are established through
requests for proposals distributed by buyers
or traders—the demand agents. If a demand
agent chooses to participate in the bilateral
market, it will first send a request for electricity with its price expectations to all the
sellers in the simulated market. In response,
a seller analyzes its own capabilities, current
availability, and past experience. The seller
must be sure that it’s feasible to deliver
energy to the buyer’s location. So, it must get
the network operator’s feedback before
www.computer.org/intelligent
reaching agreement with the demand agent.
If the seller can make an offer to the
requested parameters, it formulates a proposal and sends a message to the demand
agent. The demand agent evaluates the proposals and accepts or rejects the offers. On
the basis of the results from a negotiation
period, sellers, buyers, and traders revise
their strategies for the next period.
Hybrid market
In hybrid markets, agents must decide
whether to establish a bilateral contract either
before trying the pool or right after they learn
pool results if their bids weren’t accepted. To
make this decision, agents use past experiences and market strategies. We’ve given the
details elsewhere about how the agents handle messages in the described market types.3
Defining bids
Developing a strategic, highly profitable
offer is a fundamental issue for market participants, so the policies that each agent
implements must be analyzed carefully.
Agents take into account their past experiences and expectations about market evolution. Sellers, buyers, and traders all have
IEEE INTELLIGENT SYSTEMS
dynamic strategies to define the price at
which they’re willing to sell or buy in each
negotiation period.
These agents can also use a scenario
analysis algorithm, which we describe in
detail later. This algorithm offers more complex support for developing and implementing dynamic pricing strategies.
In terms of strategic behavior, sellers and
demand agents (buyers or traders) have similar structures and somewhat symmetrical
behavior owing to their opposite objectives.
We’ll explain the strategies from a seller’s
viewpoint. However, we’ll point out the differences when necessary.
price change produced less revenue per unit,
the seller makes a different price change.
Demand agents also use both strategies,
but with different parameters depending on
their objective of buying all the needed power
at the lower price.
For both strategies, the next period’s offer
price will be the previous period’s price
adjusted by an amount that will be more or
less than the previous price, depending on
the strategy used. The price adjustment is
determined by the same calculation:
pricei+1 = pricei ± amounti+1
amounti+1 = pricei * β +
Dynamic strategies
Agents use time-dependent strategies to
change their price during a negotiation period.
These strategies differ depending on both the
point in time when the agent starts to modify
the price and the amount it changes. Determined agents will maintain constant prices
during the negotiation period. Anxious agents
will start modifying prices early in the negotiation period by a small amount. Moderate
agents will start changing prices in the middle of the period by a small amount. Gluttonous agents will start changing prices at the
end of the period by a large amount.
Agents use behavior-dependent strategies
to adjust their price between negotiation
periods. The composed goal-directed strategy is based on two consecutive objectives—
for example, selling (or buying) all the available capacity (consumption needs) and then
increasing the profit (or reducing the payoff).
Following this strategy, sellers will lower
their price if, in the previous period, they
didn’t completely sell the available capacity.
They’ll raise the price if they sold all the
available capacity in the previous period.
Similarly, demand agents will offer a higher
price if they didn’t meet their consumption
needs in the previous period and offer less if
they succeeded in meeting their needs.
The adapted derivative-following strategy is based on a derivative-following strategy proposed by Amy Greenwald, Jeffrey
Kephart, and Gerald Tesauro.4 The working
principle is the same; however, we’ve
adapted the calculations to electricity markets. Adapted derivative following considers
the revenue the seller earned last period as a
result of the price change made between that
period and the one before it. If that price
change led to more revenue per unit, the
seller makes a similar price change. If that
NOVEMBER/DECEMBER 2003
∆i
capacity_availablei * α
∆ = capacity_available – energy_soldi
Time-dependent strategies differ
depending on the point in time
when the agent starts to modify the
price and the amount it changes.
Instead of adjusting the price each day by
a fixed percentage, the formula scales the
change by a ratio to sell all the available
capacity. The amount of change increases
with the difference between the amount of
energy the seller wanted to sell and the
amount it actually sold. β and α are scaling
factors.
We implemented these strategies and have
already obtained some results.5 The following example illustrates some differences in
how the two behavior-dependent strategies
performed.
Consider a simple scenario with few buyers and sellers. One seller is more competitive than the others, and two sellers have similar prices. So, we can analyze the strategies’
behavior in a monopoly-like situation in
periods of lower demand when only the competitive seller can sell. We can also consider
how the strategies work when the market is
competitive—in periods of higher demand.
We simulated 24 negotiation periods to analyze the effect of demand fluctuations during one day.
The adapted derivative-following strategy
www.computer.org/intelligent
obtained higher market prices than the composed goal-directed strategy (see Figure 2).
After carefully analyzing the results, we conclude that when the demand is less than the
most competitive seller’s available capacity,
the seller will lose money when using the
composed goal-directed strategy. The seller
will decrease the price and try to sell more,
which won’t be possible because of insufficient demand. However, the composed goaldirected strategy can be valuable, particularly
to increase market share when two or more
sellers are competing directly because of
similar proposed prices.
It would be interesting to develop another
strategy that combines the two described
strategies. This way, an agent could select the
most suitable strategy for each period on the
basis of market conditions. For example, if a
seller concludes that the demand is lower
than its available capacity, it will use adapted
derivative following. If it concludes that a
competitor has similar prices, it will use the
composed goal-directed strategy.
Scenario analysis algorithm
This algorithm is particularly suitable in
pool or hybrid markets, to support not only
agents’ decisions for proposing bids to the
pool but also their decisions to accept or
reject bilateral agreements (see Figure 3).
Agents can use the algorithm to analyze
different scenarios and evaluate the expected
returns. Seller and demand agents are like
players in a game, and this algorithm analyzes the possible scenarios resulting from
other agents’ past and likely future reactions.
Then, the algorithm will apply a decision
method to decide what to bid in the pool and
whether to accept a bilateral contract. Every
period includes the same steps, but the results
of the analyses from each period update the
agent’s market knowledge.
Scenarios and bid definition. Agents have
historical information about the other agents
in their market knowledge module. They can
build a profile of other agents with the
expected proposed prices, limit prices, and
capacities. With this information, the agent
constructs several scenarios and analyzes
them to determine the best way to deal with
competitors and how to bid to get a good—
or the most reliable—payoff.
To get warrantable data, each agent uses
techniques based on statistical analysis and
knowledge discovery tools, which search historical data. Usually, after a confidential
57
Market price
(cents per kilowatt-hour)
A g e n t s
a n d
M a r k e t s
4
3
2
Composed goal directed
Adapted derivative following
1
0
1
3
5
7
9
(a)
11
13
Period
15
17
19
21
23
Market price
(cents per kilowatt-hour)
4
3
2
Composed goal directed
Adapted derivative following
1
0
1
3
5
7
9
(b)
11
13
Period
15
17
19
21
23
Figure 2. The composed goal-directed and adapted derivative-following strategies’
behavior in (a) an asymmetric pool and (b) a symmetric pool.
binations possible considering the two prices
for each agent, is 2n, where n is the number
of other agents in the model.
It’s necessary to define which bids or
moves an agent or player should analyze. The
agent should analyze the incomes that result
from bidding its limit, desired prices, and
competitive prices—those that are just
slightly lower (or higher, in the demand
agent’s case) than its competitors’ prices.
Consider how a seller applies the algorithm. Let j be the seller agent doing the
analysis, capj its available capacity,
limit_ pricej its minimum acceptable price,
and desired_ pricej its expected desired
price. (A demand agent applies the algorithm similarly, except the limit price is a
maximum price instead of a minimum and
the objective is to buy all the energy it needs
at the lowest price instead of to sell at the
highest price.) Let P denote the set of all
players—sellers and demand agents—in the
market. Let ε be the smallest positive number allowed as a bidding increment. The
bids that agent j must analyze are
 bid (limit_ price j , cap j )

 bid (desired_ price j , cap j )
, ∀ i ∈P , i ≠ j ,

 bid (limit_ price i − ε , cap j )
 bid (expected_ price − ε , cap )
i
j

subject to
User
Scenario definition
Market
knowledge
limit_ pricei − ε > limit_ pricej
and
Scenarios
expected_ pricei − ε > limit_ pricej .
Bids definition
Next
period
Plays analysis
Decision method
Bid proposal in periodi
Pool
Figure 3. The scenario analysis algorithm.
period, markets disclose information about
past transactions. For example, such information from the Spanish electricity market
(OMEL) is available online at www.omel.es.
Each market player has two prices.
58
limit_ price is the minimum price if the
player is a seller and maximum price if the
player is a demand agent. expected_ price is
the previewed bid price. The number of scenarios, which result from the different comwww.computer.org/intelligent
The number of bids to analyze is 2 × n
+ 2. This number hits the maximum when
limit_ price j is smaller than the other
agents expected or the limit prices. We
will call a bid–scenario pairing a play;
then, the total number of plays to analyze
is number_of_bids × number_of_scenarios,
and the maximum value it can achieve is
(2 × n + 2)2n.
So far we’ve considered only when agents
bid their limit or expected prices. However,
an agent might bid between its limit and the
expected price, or even above it. So, if we say
that each agent might bid np prices, the number of scenarios becomes npn, and the number of plays to analyze is (np × n + 2)npn.
Even in a model with few players, the
number of plays can be high. For example,
IEEE INTELLIGENT SYSTEMS
if an agent is analyzing a scenario containing four other players and three different
prices for each, the agent must analyze a
maximum of 1,134 plays!
Furthermore, because the market is organized in several periods, an agent could
increase, decrease, or maintain its bid—three
possible actions—after each negotiation
period, increasing the number of scenarios
to analyze. So, after k periods, considering
the possible bid updates, the number of plays
to analyze becomes (np × n + 2)npn × 3(k − 1)n.
Considering the same example, after three
negotiation periods, an agent must analyze
7,440,174 plays, which is a huge number,
even for a distributed execution of the model.
However, must agents analyze every possible scenario?
Because our simulator is a decision support tool, the user should have the flexibility
to decide which and how many scenarios to
analyze. To do so, the user must define the
scenarios to simulate by specifying the price
that agents will propose:
pricei = λ × expected_ pricei
+ ϕ × limit_ pricei ,
where λ and ϕ are scaling factors that can
differ for each agent.
Suppose that the user selects λ = 0 and
ϕ = 1 for every seller and λ = 1 and ϕ = 0 for
every demand agent. This means he or she is
interested in analyzing a pessimistic scenario
(from the seller’s viewpoint). But, if the user
selects λ = 1 and ϕ = 0 for every agent, he or
she is interested in analyzing the most probable scenario.
With this formula, the user can define each
agent’s proposed prices for every scenario he
or she wants to consider. If the user defines
nc scenarios, the number of plays to analyze
is (nc × n + 2)nc.
After the agents analyze all the plays, the
algorithm will construct a matrix, obtain the
results, and apply a decision method to
decide which bid to propose.
Decision method. The matrix analysis with
the simulated plays’ results is inspired by the
game theory6,7 concepts for a pure-strategy
two-player game, assuming each player
seeks to minimize the maximum possible
loss or maximize the minimum possible
gain.
A seller—like an offensive player—will
try to maximize the minimum possible gain
by using the MaxiMin decision method. A
NOVEMBER/DECEMBER 2003
demand agent—like a defensive player—will
select the strategy with the smallest maximum payoff by using the MiniMax decision
method. In demand agents’ matrix analyses,
they select only situations in which they can
fulfil all their consumption needs. They avoid
situations in which agents will accept
reduced payoff but can’t satisfy their consumption needs completely.
After applying the decision method, the
agent selects one bid that it will propose on
the pool, unless it reaches an agreement for
a bilateral contract that’s more profitable than
the previewed pool results.
This analysis not only provides the agent
with decision support about the bid to propose in a pool but also helps improve the
The multiagent technology allied to
an object-oriented implementation
enables easy future improvements
and model enlargement.
period, if it couldn’t sell and is already bidding its limit price, it will most likely keep
its bid at the limit price. The following are
some sample rules to update demand agents’
behavior:
• If a demand agent bid is lower than the
market price and lower than its limit price,
the agent will increase its bid.
• If a demand agent bid is lower than the
market price and equal to its limit price,
the agent will maintain its bid.
Because a demand agent probably won’t
decrease its bid if it couldn’t buy in the previous period, if it couldn’t buy but is already
bidding its limit price, it will most likely keep
its bid at the limit price.
The knowledge rules update agents’ bids
in each scenario, but the number of scenarios
remains the same.
If at the end of a negotiation period the
agent concludes—by analyzing market
results—that it incorrectly evaluated other
agents’ behavior, it will fix other agents’ profiles on the basis of the calculated deviation
from the real results.
Implementation
negotiation mechanism for establishing
bilateral contracts. With this information,
the agent can evaluate a bilateral contract’s
potential benefits, compare them to the
benefits expected in a pool, and make
counterproposals.
Scenario actualization. As we stated earlier,
the analysis of each period’s results will
update the agent’s market knowledge and the
scenarios to study.
After each negotiation period, instead of
considering how other agents might increase,
decrease, or maintain their bid, agents use
knowledge rules that restrict modifications
on the basis of other agents’ expected behavior. The following are some sample rules to
update sellers’ behavior:
• If a seller bid is higher than the market
price and higher than its limit price, the
agent will decrease its bid.
• If a seller bid is higher than the market
price and equal to its limit price, the agent
will maintain its bid.
Because a seller probably won’t increase
its bid if it couldn’t sell in the previous
www.computer.org/intelligent
We developed MASCEM in the Open Agent
Architecture (www.ai.sri.com/~oaa) and
Java. OAA, developed at SRI International,
is a framework for integrating a community
of heterogeneous software agents in a distributed environment. OAA is structured to
minimize the effort of creating new agents
written in various languages and operating
platforms, to encourage the reuse of existing
agents, and to allow the creation of dynamic,
flexible agent communities.
The OAA’s Interagent Communication
Language is the interface and communication language that all agents share, no matter
which machine they run on or language
they’re programmed in. Because the OAA
framework isn’t specifically devoted to
developing simulations, we made it suitable
by extensions—for example, we included a
clock to introduce the simulation’s time evolution mechanism.
We implemented each agent in Java as a
Java thread. The model can be distributed
over a computer network, which makes it
easier to increase the number of simulation
runs for scenarios with many agents.
Figure 4 shows the prototype interface for
a simple scenario with three seller agents,
two buyer agents, and a trader.
59
A g e n t s
T h e
a n d
M a r k e t s
A u t h o r s
Isabel Praça is an assis-
tant professor of computer engineering at the
Polytechnic Institute
of Porto’s Institute of
Engineering and is preparing her PhD thesis
on multiagent simulation of competitive electricity markets. Her research interests are in
multiagent simulation and decision support systems. She received her MSc in electrical and
computer engineering from the University of
Porto. Contact her at Inst. Superior de Engenharia do Porto, Dept. de Engenharia Informática,
Rua Dr. António Bernardino de Almeida, 4200072 Porto, Portugal; [email protected].
Carlos Ramos is a
coordinator professor
of computer engineering and director of the
Knowledge Engineering and Decision Support Research Group at
the Polytechnic Institute of Porto’s Institute
of Engineering. His research interests are artificial intelligence and decision support systems.
He received his PhD in electrical engineering
from the University of Porto. Contact him at
Inst. Superior de Engenharia do Porto, Dept.
de Engenharia Informática, Rua Dr. António
Bernardino de Almeida, 4200-072 Porto, Portugal; [email protected].
Zita Vale is a coordinator professor of electrical engineering at
the Polytechnic Institute of Porto’s Institute
of Engineering. Her
research interests are
decision support and
artificial intelligence.
She received her PhD in electrical engineering
from the University of Porto. She is a member
of the IEEE, IEE, and ACM. Contact her at Inst.
Superior de Engenharia do Porto, Dept. de
Engenharia Electrotécnica, Rua Dr. António
Bernardino de Almeida, 4200-072 Porto, Portugal; [email protected].
Manuel Cordeiro is a
full professor at the
University of Trás-osMontes and Alto Douro.
His research interests are
rational energy use,
grounding systems, and
power systems applications. He received his
PhD in electrical engineering from the University of
Trás-os-Montes and Alto Douro. Contact him at
Univ. de Trás-os-Montes e Alto Douro, Engenharias
II, 5000-911 Vila Real, Portugal; [email protected].
60
Figure 4. The prototype MASCEM interface.
B
ecause of the growing number of entities and the redefined rules in newly
competitive electricity markets, MASCEM
seems to be a valuable framework for studying market evolution. The multiagent technology allied to an object-oriented implementation enables easy future improvements
and model enlargement. This is important,
considering markets are still very much in
the evolutionary process.
Market participants’ strategic behavior is
very significant in the context of competition. Although we implemented some valuable, promising strategies, we must enlarge
the portfolio of agents’ strategies and behaviors. For example, we’ll implement and test
strategies based on learning algorithms, such
as Q-learning.
Developing targeted studies to analyze
specific markets, such as the emerging market between Portugal and Spain known as the
Iberian Market, will also be an important step
toward validating the profitability of using
tools such as MASCEM.
www.computer.org/intelligent
References
1. I. Praça et al., “Multi-Agent Simulation of
Electricity Markets,” Proc. 2nd Workshop
Agent-Based Simulation, Soc. for Computer
Simulation, 2001, pp. 89–94.
2. M. Shahidehpour, H. Yamin, and Z. Li, Market Operations in Electric Power Systems:
Forecasting, Scheduling and Risk Management, John Wiley & Sons, 2002.
3. I. Praça et al., “Intelligent Agents for Negotiation and Game-Based Decision Support in
Electricity Markets,” Proc. 12th Intelligent
System Application to Power Systems Conf.
(ISAP 03), 2003.
4. A. Greenwald, J. Kephart, and G. Tesauro,
“Strategic Pricebots Dynamics,” Proc. 1st
ACM Conf. Electronic Commerce, ACM
Press, 1999, pp. 58–67.
5. I. Praça et al., “Agent Strategies in MultiPeriod Negotiations for Electricity Markets,”
Proc. 2nd IASTED Int’l Conf. Artificial Intelligence and Applications, ACTA Press, 2002,
pp. 411–416.
6. J. Von Neumann and O. Morgenstern, Theory
of Games and Economic Behavior, Princeton
Univ. Press, 1947.
7. D. Fudenberg and J. Tirole, Game Theory,
MIT Press, 1991.
IEEE INTELLIGENT SYSTEMS
PURPOSE The IEEE Computer Society is the
PUBLICATIONS AND ACTIVITIES
world’s largest association of computing
professionals, and is the leading provider of
technical information in the field.
Computer. An authoritative, easy-to-read
magazine containing tutorial and in-depth
articles on topics across the computer field,
plus news, conferences, calendar, industry
trends, and product reviews.
MEMBERSHIP Members receive the
monthly magazine COM PUTER , discounts,
and opportunities to serve (all activities
are led by volunteer members). Membership is open to all IEEE members, affiliate
society members, and others interested in
the computer field.
Periodicals. The society publishes 12
AVAILABLE INFORMATION
To obtain more information on any of the
following, contact the Publications Office:
• Membership applications
COMPUTER SOCIETY WEB SITE
The IEEE Computer Society’s Web site, at
http://computer.org, offers information
and samples from the society’s publications
and conferences, as well as a broad range
of information about technical committees,
standards, student activities, and more.
Members experiencing
problems—magazine delivery, membership status, or unresolved complaints—
may write to the ombudsman at the
Publications Office or send and e-mail
to [email protected].
OMBUDSMAN
CHAPTERS Regular and student chapters
worldwide provide the opportunity to
interact with colleagues, hear technical
experts, and serve the local professional
community.
• Publications catalog
• Draft standards and order forms
• Technical committee list
• Technical committee application
• Chapter start-up procedures
• Student scholarship information
• Volunteer leaders/staff directory
• IEEE senior member grade application (requires 10 years practice and significant performance in five of those 10)
To check membership status or report a
change of address, call the IEEE toll-free
number, +1 800 678 4333. Direct all other
Computer Society-related questions to the
Publications Office.
E X E C U T I V E
President:
STEPHEN L. DIAMOND*
Picosoft, Inc.
P.O.Box 5032
San Mateo, CA 94402
Phone: +1 650 570 6060
Fax: +1 650 345 1254
[email protected]
President-Elect:
CARL K. CHANG*
Past President:
WILLIS. K. KING*
VP, Educational Activities:
DEBORAH K. SCHERRER (1ST VP)*
VP, Conferences and Tutorials:
CHRISTINA SCHOBER*
VP, Chapters Activities:
MURALI VARANASI†
VP, Publications:
RANGACHAR KASTURI†
BOARD OF GOVERNORS
Term Expiring 2003: Fiorenza C. AlbertHoward, Manfred Broy, Alan Clements, Richard
A. Kemmerer, Susan A. Mengel, James W. Moore,
Christina M. Schober
Term Expiring 2004: Jean M. Bacon, Ricardo
Baeza-Yates, Deborah M. Cooper, George V.
Cybenko, Haruhisha Ichikawa, Lowell G. Johnson,
Thomas W. Williams
Term Expiring 2005: Oscar N. Garcia, Mark A.
Grant, Michel Israel, Stephen B. Seidman, Kathleen
M. Swigger, Makoto Takizawa, Michael R. Williams
Next Board Meeting: 22 Nov. 2003, Tampa, FL
EXECUTIVE
Conference Proceedings, Tutorial
Texts, Standards Documents.
The Computer Society Press publishes
more than 160 titles every year.
Standards Working Groups. More
than 200 groups produce IEEE standards
used throughout the industrial world.
Technical Committees. Thirty TCs publish newsletters, provide interaction with
peers in specialty areas, and directly influence standards, conferences, and education.
Conferences/Education. The society
holds about 100 conferences each year
and sponsors many educational activities,
including computing science accreditation.
C O M M I T T E E
VP, Standards Activities:
JAMES W. MOORE†
VP, Technical Activities:
YERVANT ZORIAN†
Secretary:
OSCAR N. GARCIA*
Treasurer:
WOLFGANG K. GILOI* (2ND VP)
2002–2003 IEEE Division V Director:
GUYLAINE M. POLLOCK†
COMPUTER SOCIETY O F F I C E S
IEEE
2003–2004 IEEE Division VIII Director:
JAMES D. ISAAK†
2003 IEEE Division V Director-Elect:
GENE H. HOFFNAGLE
Computer Editor in Chief:
DORIS L. CARVER†
Executive Director:
DAVID W. HENNAGE†
* voting member of the Board of Governors
† nonvoting member of the Board of
Governors
OFFICERS
Headquarters Office
1730 Massachusetts Ave. NW
Washington, DC 20036-1992
Phone: +1 202 371 0101 • Fax: +1 202 728 9614
E-mail: [email protected]
President:
MICHAEL S. ADLER
Publications Office
10662 Los Vaqueros Cir., PO Box 3014
Los Alamitos, CA 90720-1314
Phone:+1 714 821 8380
E-mail: [email protected]
Membership and Publication Orders:
Phone: +1 800 272 6657 Fax: +1 714 821 4641
E-mail: [email protected]
Executive Director:
DANIEL J. SENESE
STAFF
Executive Director: DAVID W. HENNAGE
Assoc. Executive Director:
ANNE MARIE KELLY
Publisher: ANGELA BURGESS
Assistant Publisher: DICK PRICE
Director, Finance & Administration:
VIOLET S. DOAN
Director, Information Technology & Services:
ROBERT CARE
Manager, Research & Planning: JOHN C. KEATON
magazines and 10 research transactions.
Refer to membership application or request
information as noted at left.
Asia/Pacific Office
Watanabe Building
1-4-2 Minami-Aoyama,Minato-ku,
Tokyo107-0062, Japan
Phone: +81 3 3408 3118 • Fax: +81 3 3408 3553
E-mail: [email protected]
President-Elect:
ARTHUR W. WINSTON
Past President:
RAYMOND D. FINDLAY
Secretary:
LEVENT ONURAL
Treasurer:
PEDRO A. RAY
VP, Educational Activities:
JAMES M. TIEN
VP, Publications Activities:
MICHAEL R. LIGHTNER
VP, Regional Activities:
W. CLEON ANDERSON
VP, Standards Association:
GERALD H. PETERSON
VP, Technical Activities:
LEAH JAMIESON
IEEE Division VIII Director
JAMES D. ISAAK
President, IEEE-USA:
JAMES V. LEONARD