read_me.txt

使用遗传算法计算电力市场投标程序

buyers and sellers a well as the number of buyers and 
sellers and the transmission capacities connecting them.

The EXIT command allows you to leave the simulator.  You 
can also leave by clicking the close box in the upper right 
corner of the window or by using the generic Windows( menu 
exit function.

The EDIT menu contains the following entries:  

BUYERS, SELLERS, and TRANSCAPS.

The BUYERS commands both bring up dialogs for editing a set 
of agents.  There are three sliders that can be used with 
their accompanying "Set All" buttons to give all buyers the 
same megawatts to buy, revenue per megawatt, and fixed 
cost.  The "Done" button exits the editor.  The "Reset 
Genes" button resets the buyers genes.  The "Random Genes" 
button generates random genes if they are needed for 
experimentation. The rows of matrix data in each dialog 
describe the individual buyers and may be clicked and 
edited for each individual.

The SELLERS command brings up dialogs for editing a set of 
agents.  There are three sliders that can be used with 
their accompanying "Set All" buttons to give all sellers 
the same megawatts to sell, cost per megawatt, and fixed 
cost.  The "Done" button exits the editor.  The "Reset 
Genes" button resets the sellers genes.  The "Random Genes" 
button generates random genes if they are needed for 
experimentation.  The rows of matrix data in each dialog 
describe the individual sellers and may be clicked and 
edited for each individual.  Entering patently absurd 
values will cause problems and such practices are 
discouraged.

The TRANSCAPS command summons up the dialog that specifies 
the number of buyers and sellers and the transmission 
capacities linking them.  The number of buyers and sellers 
are controlled by the sliders in the lower left hand 
corner.  The transmission capacities can be directly edited 
by clicking on them.  It is also possible to set them all 
at once by using the uniform slider and button on the lower 
right.  The "Done" button exits the dialog.  The reset 
restores all values to their default - six buyers and 
sellers and a uniform transmission capacity of 10.

The GENETICS menu consists of the following buttons:

ALL, RESET GENES, RESET PARAMS, and RANDOM SEED.

The ALL command summons up four dialogs, described below, 
for controlling crossover type, number of generations, 
mutation rate, and selection technique.  The state of all 
four of these dialogs is saved and loaded when the save and 
load commands of the file menu are used.

CROSSOVER is the blending of two existing genes to produce 
a gene, for use in searching for effective bidding 
strategies.  A radio button group allows the user to select 
one point, two points, uniform, or no crossover.  These 
terms are explained in Section 2.

GENERATION consists of a computation of the bidding and 
profit mechanics, followed by the creation of new 
strategies by selection, crossover, and mutation.  This 
menu entry gives access to the generation slider that 
permits the user to select the number of generations to be 
executed and displayed.

MUTATION represents the changing of bits in the gene in a 
random fashion.  This is also part of the search for 
locally better bidding strategies.  A slider allows the 
user to control the per-bit mutation probability.  When 
mutating a gene the genetic algorithm check a random number 
for each bit to see if that bit should be flipped.  
Mutation is placed in context in Section 2.

SELECTION is the process of picking which genes will be 
allowed to replace others.  For the genetic algorithm to 
perform well superior strategies must, on average, replace 
inferior ones.  There are a number of ways to do this (see 
Section 2 and a radio button group lets the user select 
between PROPORTIONAL SELECTION, RANK SELECTION, and 
ROULETTE SELECTION.  The constant and ratio for 
proportional selection are controlled by sliders at the 
bottom of the dialog.

RESET GENES command sets the buyers genes to whatever gene 
corresponds to a bid equaling the revenue per megawatt that 
was used the last time a uniform revenue per megawatt was 
chosen.  This is done to the limit of the resolution of the 
genetic encoding and will usually be a few pennies low.  
The Reset Genes command also sets the sellers genes to all 
zeros.  These are the extreme, economically rational 
values, most generous to an individuals opponent, from 
which the simulator improves.  The buyer and seller 
dialogs, described above can be used to randomize or edit 
individual genes.

The RESET PARAMS command restores all of the genetic 
parameters, crossover type, mutation rate, generations, and 
selection type, to their default values.

The RANDOM SEED command allows the user to choose how the 
random number generator used by the simulator is seeded.  
If the "take from system clock" checkbox is enabled then 
the random numbers will depend on the time of day to the 
thousandth of a second; behavior will change with each 
repetition.  Disabling this checkbox causes the simulator 
to use the (32-bit signed) random number seed in the edit 
box.  Use of a user selected seed permits repetition of 
experiments.  This is the only way to debug a genetic 
algorithm!

The Display menu consists of the following buttons:

ALL, BIDS, LOGO, MAX. PROFITS, NOTHING, PROFITS, and 
TRANSCAPS.

The ALL command displays all four types of data taken 
during simulation: average bids, maximum profits, average 
profits, and use of transmission capacity.  The individual 
displays are a little small.  Larger versions of each 
individual data set can be obtained by using the commands 
of the same name.

The BIDS, MAX. PROFITS, PROFITS, and TRANSCAPS commands 
give a full screen display of the data sets they are named 
for.  The units of the display are generations of 
simulation along the horizontal axis and are drawn from the 
values in the Buyer, Seller, and Transcap Dialogs described 
above along the vertical axis.  They are automatically 
resized and so may not compare well visually across 
substantially different numbers of agents.

The LOGO and NOTHING commands display the program's start 
up screen (the logo) and an empty screen respectively.  
These are not important or useful commands - but the 
screens in the simulator are a little busy and a plain 
screen can be pleasant when figuring out what to do next.

The RUN menu consists of a single button: 

GA.

The GA command runs the genetic algorithm.  You should see 
(depending on the speed of your computer and the number of 
buyers and sellers) a slider move from left to right.  When 
the slider disappears four small data displays appear, 
showing bids, average profits, maximum profits, and 
transmission capacity used.

Each of the six tutorial commands runs a file viewer 
containing a tutorial about some part of the simulator.  
The topic of each tutorial is given in its first screen.  
The tutorial screens are also included at the end of this 
documentation.  The topics of the tutorials are:

1]  Introduction:  running the genetic algorithm and 
viewing data.
2]  Data Modification:  editing buyers, sellers, and 
transmission capacities.
3]  Genetic Options:  controlling the genetic algorithm 
parameters.
4]  File Usage:  file handling.
5]  Random numbers:  seeding options for the pseudo-random 
number generator.
6]  Transport Limits:  editing and use of the transmission 
capacity matrix.

The tutorial information as provided with the diskette is 
shown in Appendix A.


EXPERIMENTS

There are three cases, which are consistently used to 
demonstrate the algorithm.  The first test case is to set 
all of the buyers and sellers equal, raise the 
transportation limits above the necessary flows, set the 
mutation rate to zero (0), and run the simulator.  The 
result is quickly found to be the optimal supply-demand 
equilibrium point if a central pool knew all data.  The 
surplus profit is reduced to zero and a single equilibrium 
is found.  If the mutation rate is now set to a reasonable 
value (5%-15%), then there will be excursions of sellers 
into the buyer's profit.  The sellers are more aggressive 
due to their GA structure.  This would be a reasonable 
model for a market environment where the GENCOs are 
unregulated and the DISTCOs are regulated.  

A second case is to select one or more sellers to be 
expensive when compared to the remaining sellers.  If the 
demand for all of the buyers do not use the entire lessor 
cost generation, then the more expensive companies do not 
find a stable profit level to operate and to survive.  This 
would compare to a GENCO, which should be merged or 
retired.

A third case is to introduce a restriction between the 
sellers and the buyers.  Just a single restriction will 
lead to very violent activity as the geographically 
restricted buyers and sellers attempt to find a profit 
level to survive.  It would be more realistic to include a 
common data knowledge base to emulate OASIS or any 
financial report on what the competition is performing.