Background
The electric power industry is undergoing rapid and significant change with the advent of deregulation. Electric power marketers have emerged, and wholesale electric customers are learning to "shop around" for the best electric suppliers. At the same time, the industry is adopting Internet-based electronic scheduling and trading systems that provide open access to all transmission services information for all market participants. With increased competition, the utilities are discovering that improved productivity in power generation and transmission takes on renewed importance. New tools and technologies are required to allow the industry to grow and thrive in this new era of deregulation. In particular, new methods of controlling, scheduling, buying and selling electricity are needed.

Research Objectives
Acronymics Inc. personnel and Alternative Energy Systems Consultants (AESC) have conducted extensive research into these problems. Clients include the US Department of Energy, the California Energy Commission and Electric Power Research Institute (EPRI) in Palo Alto, CA. The EPRI project focused on the develop-ment of intelligent agent technology and application of that technology to the electric power market. We have demonstrated networks of communicating and cooperating intelligent software agents can be used to implement complex distributed systems; distributed systems that implement electronic marketplaces. As part of this project, we have shown how collections of agents (agencies) can be used to buy and sell electricity and participate in the electronic marketplace. Agents can act as buyers or sellers, auctioneers or bidders and representatives for generators or consumers of electricity.

Demonstration Overview
As part of this research, Acronymics, Inc. personnel and AESC developed a prototype agent-based auction system to illustrate the use of agent technology in an electronic commerce application. This demonstration is designed to show how agents can be used to implement an electronic marketplace. This demonstration is not meant to be a prototype of an actual marketplace. Because of budget and schedule constraints, the scope of this demonstration is quite limited. The demonstration shows how intelligent software agents acting as buyers and sellers of electric power can be used to implement an on-line auction.

Demonstration Agents
The demonstration marketplace is shown in Figure 1. The market is comprised of three different kinds of agents. These are the Buyer/Seller Agents, the Facilitator Agent and an Experiment Control Agent. The following paragraphs provide a brief description of the function of each of these agents.

Buyer/Seller Agent (BSA)
The Buyer/Seller agents (BSA) buy and sell electric power. They represent either electric power producers (generators) or electric power consumers (loads). These agents are the primary participants in the marketplace and do all of the buying and selling of electricity.

A BSA can take on a number of different roles. Agents can, at any instant in time, be either buyers or sellers of electricity; they can also either be conducting an auction (to buy or sell) or they can participate in an auction as a bidder (to buy or sell).

Each BSA exhibits unique and individual behavior determined by its own unique economic and behavioral model. There are three components of each agent's behavior. These are:

• buying/selling strategy
• market supply/demand characteristics
• operational power quantity requirements

Strategy. Each BSA has a unique buying and selling strategy. BSAs can exhibit anxious buying and selling behavior, cool-headed behavior or greedy behavior. See Figure 2 for an illustration of different buying and selling strategies.

Agent Buying and Selling Strategies

Market/Supply Demand. Each BSA has its own unique supply or demand requirements and unique operating policies for satisfying supply/demand requirements. These policies dictate how much each BSA is willing to spend/charge for power.

Operational Requirements. Each BSA represents a supplier or consumer of electricity. Each supplier or consumer has it own unique power requirements that vary over time. While market price may dictate agent buying/selling behavior to some extent, agents must still ensure that adequate amounts of power are bought and sold to sustain operations.

These three factors are used to compute supply and demand curves that determine the amount of power that each agent will require based on the price of that power. As prices go up, agents will be willing to consume less power and be willing to supply more power. As prices fall, agents will be willing to consume more power or sell less power. A supply curve describes the desired selling price for a quantity of electricity to be delivered at a given date and time. A demand curve describes the required buying price for buying a given quantity of electricity at a given date and time.

Other Useful Agents

Facilitator Agent
The facilitator agent provides a number of unique services to the electronic marketplace. The facilitator keeps tracks of all BSAs that are participating in the marketplace. Each BSA communicates with the Facilitator and indicates whether it wishes to buy or sell electricity and whether it wants to be a bidder or auctioneer in the marketplace. Thus, the Facilitator acts as a broker or matchmaker connecting buying agents with selling agents and auctioneering agents with bidder agents. Since the demonstration system is quite small with only a few agents, no facilitator agent is necessary. Each agent has a priori knowledge of the existence and location of all other agents.

Grid Interface Agent
The Grid Interface Agent (GIA) performs two major functions in the demonstration system. First, the GIA provides a mechanism for conveniently simulating the power industry infrastructure. In an operational system, this infrastructure (power supply and demand) will dictate the behavior of the BSAs. The GIA provides a convenient mechanism for setting up and controlling the operation of the individual BSAs.

In a production system, the GIA would be replaced by the institutional infrastructure that specifies the behavior of each BSA. That is, the buying and selling prices will normally be driven by economic and physical factors (e.g., how much power is available to sell and how much surplus generation capability is available for use). However, for the demonstration, we need a way of specifying the behavior of each of the BSAs. The GIA provides a convenient interface for doing this.

The GIA also provides a convenient way of setting up and controlling each BSA and examining the results of BSA activities (buying and selling). Thus, the GIA collects data from all market sessions and displays it in a convenient way (i.e., using the graphical user interface on the GIA). For this simple demonstration, the Experiment Control Agent provides the functionality that would be found in a GIA in a production auction system.

Experiment Control Agent
The Experiment Control Agent provides a graphical user interface to control the simulated auction session and view the results as they are generated. Figure 3 illustrates the user interface for the experiment control agent. The Experiment Control Agent is used to set up the agency of Buyer/Seller Agents and controls the overall timing of the auction. The Experiment Control Agent does not participate in the buying and selling of electric power. Its primary purpose is to allow an auction to be conducted as a demonstration rather than in real time (which would take one week). The Experiment Control Agent sends simulated times to the buyers and sellers and collects auction results for logging and display.

Experiment Control Agent User Interface

The AgencyViewer tool is a graphical tool that allows the user to view communication messages flowing between the agents. The AgencyViewer tool is part of the AgentBuilder product. An icon represents each agent and lines are drawn between the icons to represent each communication message sent between agents. . When the AgencyViewer is started a window with eleven agent icons will appear. The are five bidder/buyers (BB1 - BB5), five sellers/auctioneers (SA1 - SA5) and one experiment control agent (ExpCtl). Figure 4 illustrates this tool. For more information on AgentBuilder and the AgencyViewer tool review the AgentBuilder documentation. This documentation is available for download.

The AgencyViewer Examining the Auction Agency

The Market Place
The market consists of a combination of buyer and seller agents taking on roles as either bidders or auctioneers. The agents are each assigned a role (as buyer or seller and auctioneer or bidder) by the demonstration operator using the Experiment Control Agent. The agents then conduct an auction buying and selling electricity in accordance with the economic needs and buying and selling strategies of the individual agents.

Demonstration Scenario Description
An agent initiates an auction by advertising that it is going to conduct an auction and specifying the time of the auction. All bidders interested in participating in the auction can then join the auction and submit bids. Each auctioning agent can be viewed as conducting an auction that is selling/buying 24 items - power supply or demand for one specified hour on a given day in the future.

The auctioning agent will offer a quantity Q of electricity at an asking price AP for delivery at hour H on delivery day D. The agent will also specify when the next round of the auction will be held for selling the remaining power not sold in the current round as well as the price adjustment PA that will be applied to the remaining power asking price in the next round. Each auctioning agent will also maintain a reserve price RP that specifies the minimum/maximum price that the agent will accept before refusing any further transactions at a lower/higher price. The RP is known only to the auctioning agent and is not made available to bidding agents. Note that an auctioning agent may decide to reintroduce this block of capacity back into the marketplace at some later time.

Agents that want to buy/sell power from/to the auctioning agent specify the quantity that they are willing to buy/sell at the specified price. The auctioneer then forms a contract with the bidder to deliver that power (subject to collision constraints). Agents are encouraged to buy during a particular round because they cannot be sure that they can get the power they require in a later round if it is purchased in this round by some other agents. While bidder agents may wish to defer contracting with the auctioneer, they do so at the risk of being unable to acquire the amount of power that they require.

Figure 5 illustrates a typical auction. For the demonstration, the initial round is for delivery of power at the specified hour seven days in the future. The next round is for delivery of power at that same hour six days in the future. If bidding continues to the last round, then this round is for sale of power to be delivered on the next day. Note in this particular example that the price profile shows that the agent will reach its reserve price before all seven rounds of the auction is completed. Since bidders don't know when the reserve price will be reached and the remaining power withdrawn from the market, they are encouraged to bid early in the round.

The Auction Process

Summary
This research has shown that networks of intelligent agents provide a powerful problem-solving mechanism that is ideally suited for use in the electric power industry. Agents are ideally suited for use in network applications. The new restructured and unregulated power industry will utilize network communications and the Internet extensively. We have shown that agents provide an excellent mechanism for implementing information systems for buying, selling and scheduling electric power. We have shown how software agents can be used to represent the interest of various stakeholders involved in buying and selling electric power.

Running the Demonstration System

Each agent has its own goals, objectives, cost and profit objectives. Further, agents work together to form markets with other agents buying and selling electric power and change their role as buyer or seller as their electric power requirements change. A white paper describing the design of a prototypical electronic marketplace for buying and selling electric power is available for download. In addition, a simple demonstration system (coming soon) that shows the operation of the agents and the marketplace is available for download.

In summary this research has shown:

• Agents can be used to implement electronic marketplaces and electronic auctions.
• An individual agent can adopt the goals and intentions of its stakeholder.
• A central market authority is not necessary. Agents can dynamically form their own marketplaces meeting their individual needs.
• The distributed, multiple agent approach is significantly more useful than a central server, since individual agent behaviors are under control of the stakeholder and not subject to the rules and constraints of a central authority.

Additional Information
Additional information concerning the work described in this paper can be found at a number of sources.

Formal Report. This work is more fully documented in a detailed technical report TR-113366 titled "Prototype Intelligent Software Agents for Trading Electricity: Competitive/Cooperative Power Scheduling in an Electronic Marketplace." Please contact EPRI directly for a copy of this document.
White Paper. A white paper describes this research and is available for download in pdf format.
AgentBuilder. The AgentBuilder Pro product developed and marketed by Acronymics, Inc. was used in development of the agents and agency described in this white paper.

If you have additional questions, please contact us at informationAgent@agentbuilder.com with Subject line "EPRI."

For questions/comments about this website, please contact informationAgent@agentbuilder.com
Last modified: November 15, 2011
AgentBuilder is a registered trademark of Acronymics, Inc.
All other product names referenced are trademarks of their respective companies.
Copyright © 2004 - 2011 Acronymics, Inc.
Privacy Policy