read_me.txt

使用遗传算法计算电力市场投标程序

The complete report may be obtained by contacting the 
Electric Power Research Institute.

Computer Simulation of Adaptive Agents for an Electric 
Power Auction
(Genetic Bidding for Electric Markets)

Project TR-107975
April 15, 1997

Prepared by
EPMT, Inc.
3442 Southdale Drive
Ames, Iowa  50010

Project Manager/Author
Gerald B. Sheble'

Prepared for
Electric Power Research Institute
3412 Hillview Avenue
Palo Alto, California  94304

EPRI Project Manager
Martin Wildberger
Office of Exploratory and Applied Research


DISCLAIMER OF WARRANTIES AND LIMITATION OF LIABILITIES
THIS REPORT WAS PREPARED BY THE ORGANIZATION(S) NAMED BELOW 
AS AN ACCOUNT OF WORK SPONSORED OR COSPONSORED BY THE 
ELECTRIC POWER RESEARCH INSTITUTE, INC. (EPRI). NEITHER 
EPRI, ANY MEMBER OF EPRI, ANY COSPONSOR, THE 
ORGANIZATION(S) BELOW, NOR ANY PERSON ACTING ON BEHALF OF 
ANY OF THEM:

(A) MAKES ANY WARRANTY OR REPRESENTATION WHATSOEVER, 
EXPRESS OR IMPLIED, (I) WITH RESPECT TO THE USE OF ANY 
INFORMATION, APPARATUS, METHOD, PROCESS, OR SIMILAR ITEM 
DISCLOSED IN THIS REPORT, INCLUDING MERCHANTABILITY AND 
FITNESS FOR A PARTICULAR PURPOSE, OR (II) THAT SUCH USE 
DOES NOT INFRINGE ON OR INTERFERE WITH PRIVATELY OWNED 
RIGHTS, INCLUDING ANY PARTY'S INTELLECTUAL PROPERTY, OR 
(III) THAT THIS REPORT IS SUITABLE TO ANY PARTICULAR USER'S 
CIRCUMSTANCE; OR

(B) ASSUMES RESPONSIBILITY FOR ANY DAMAGES OR OTHER 
LIABILITY WHATSOEVER (INCLUDING ANY CONSEQUENTIAL DAMAGES, 
EVEN IF EPRI OR ANY EPRI REPRESENTATIVE HAS BEEN ADVISED OF 
THE POSSIBILITY OF SUCH DAMAGES) RESULTING FROM YOUR 
SELECTION OR USE OF THIS REPORT OR ANY INFORMATION, 
APPARATUS, METHOD, PROCESS, OR SIMILAR ITEM DISCLOSED IN 
THIS REPORT.
ORGANIZATION(S) THAT PREPARED THIS REPORT

NAME OF COMPANY:   EPMT, Inc.

ORDERING INFORMATION

Requests for copies of this report should be directed to 
the EPRI Distribution Center, 207 Coggins Drive, P.O. Box 
23205, Pleasant Hill, CA 94523, [phone]. There is no 
charge for reports requested by EPRI member utilities.
Electric Power Research Institute and EPRI are registered 
service marks of Electric Power Research Institute, Inc.

Copyright (c) 1997 Electric Power Research Institute, Inc. 
All rights reserved.

Excerpts from report follow.

INTRODUCTION
 
In this program, we use adaptive agents used to simulate a 
double auction market for buying and selling electric 
energy.  The adaptation of the agents is accomplished with 
a basic genetic algorithm.  This algorithm encourages each 
agent to locally maximize their profit in an environment 
where both buyers and sellers are changing their bids.  In 
such an environment the market acts, via the local 
optimizations, to find the market price in an efficient 
fashion.  This location of the market price happens without 
any special knowledge or internal modeling of the market by 
the agents.  The mere fact that successful agents pass 
their behavior along, with exploratory modifications, to 
unsuccessful agents is enough to move the market toward 
mutually agreeable prices in a relatively short time.  
Economically, the market is found to be a stable auction 
mechanism.  However, the market is not efficient in finding 
a solution where one agent is unfairly splitting the 
surplus profit with another agent.

The agents used in this simulation live in an environment 
where multiple power producers and consumers are engaging 
in competitive bidding for electric power.  The feature of 
the agents that adapts is the price for which they are 
willing to buy or sell power.  The production, demand, 
variable and fixed costs, and transmission costs all remain 
fixed.  Once the agents have made their bids for the best 
buy and sell offers, matches of buyers and sellers are made 
as long as offered price exceeds asked price, and power is 
sold (contracted) at a price at the midpoint between the 
asked and offered prices.  This matching continues until 
all possible transactions are paired for which each agent 
gets the amount bid.  There are two components of the 
simulator: market structure and rules of engagement, and 
genetic algorithm optimization of agent's bid.



SIMULATOR MARKET STRUCTURE

The market structure consists of two types of players: 
buyers and sellers in addition to the central broker.  Each 
seller is modeled as a corporation with variable and fixed 
costs of production.  Each buyer is modeled as a 
corporation with fixed and variable costs of delivery.  
Each buyer and seller has a fixed amount to buy or sell.  
Each buyer and seller is given a network location, which is 
used to determine the transportation capability from each 
buyer to each seller individually.

The broker matches the bids from each buyer to each seller 
by ordering the bids inversely by type of player.  Thus the 
highest buyer is matched with the lowest seller.  The 
broker then verifies that the transaction can occur based 
on the remaining network capability between these players.  
If sufficient capacity exists, then the transaction is 
committed, the remaining transportation matrix is updated 
to reflect the actual flow based on the contract, and the 
next buyer and seller is matched.  A buyer or seller may 
have to engage in more than one transaction if the amount 
bid is not equal or if the network restricts the flow.  The 
broker always commits transactions to sell all bid amounts 
unless restricted by the network.  The broker continues to 
match bids until the surplus profit is zero.  The surplus 
profit is the difference between the buyers bid and the 
sellers bid.  The transaction price is always the 
difference between the two bids, divided by two.


RUNNING THE PROGRAM 

The EPRI Market Simulator simulates a double auction market 
in which buyers and sellers of electric power (termed 
agents) make transactions.  The simulation of the evolution 
over the course of bidding of the agents is accomplished by 
means of a genetic algorithm the exact nature of which may 
be specified by the user.  The simulator computes and 
displays average bids, average profits, maximum profits, 
and use of transmission capacity during the course of 10-
200 rounds of bidding.  The user is permitted to specify 
the number of buyers and sellers, the transmission 
capacity, and the individual economic constraints such as 
quotas, fixed costs, and marginal costs for each agent.  
The program can save and load entire market setups 
including the parameters of each agent and the transmission 
network.  The current version of the simulator is intended 
as a demonstration tool.

The simulator is menu driven with Windows( dialogs for 
exchanging information with the user.  The buttons have 
been designed to speed access of the data.  The follow 
description will define the buttons at each level and then 
describe the individual commands associated with the 
buttons at that level.

The main menu consists of the following buttons:

FILE, EDIT, GENETICS, DISPLAY, RUN, TUTORIAL.

The FILE menu contains the load and save commands for 
simulation data sets as well as the exit function to leave 
the simulator.

The EDIT menu contains commands for pulling up the dialogs 
to view or modify the buyers, sellers, or the transmission 
capacities.

The GENETICS menu pulls up the dialogs that set controls 
for the genetic algorithm.  It can also reset the agent's 
genes and the control parameters for the genetic algorithm 
to their default values.  The controls for random number 
seeding are also accessed through the genetics menu.

The DISPLAY menu has commands that permit each of the four 
types of data gathered by the simulator to be displayed, 
individually or together.  It also permits clearing of the 
screen or display of the starting screen (logo).

The RUN menu contains the command for running the genetic 
algorithm.

The TUTORIAL menu runs a separate file viewer program to 
display a number of tutorials.  These cover the controls of 
the program and it is practical to learn the program on 
line by using these tutorials.

The FILE menu system contains the following entries:

LOAD, SAVE AND EXIT.

The LOAD and SAVE commands save all the information 
contained in the Buyers, Sellers, and Transmission 
capacities (Transcaps) dialogs as well as state of the 
genetic algorithm control dialog.  This does not include 
the information on bids, profits, and use of transmission 
capacity generated when the genetic algorithm is run.  It 
does include the defining statistics and genes of the