CN110310184B - Power generation market bidding simulation method based on multi-agent game with memory function - Google Patents

Power generation market bidding simulation method based on multi-agent game with memory function Download PDF

Info

Publication number
CN110310184B
CN110310184B CN201910615663.7A CN201910615663A CN110310184B CN 110310184 B CN110310184 B CN 110310184B CN 201910615663 A CN201910615663 A CN 201910615663A CN 110310184 B CN110310184 B CN 110310184B
Authority
CN
China
Prior art keywords
agent
generator
bidding
intelligent
market
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Active
Application number
CN201910615663.7A
Other languages
Chinese (zh)
Other versions
CN110310184A (en
Inventor
黄仙
张军
李树松
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
North China Electric Power University
Original Assignee
North China Electric Power University
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by North China Electric Power University filed Critical North China Electric Power University
Priority to CN201910615663.7A priority Critical patent/CN110310184B/en
Publication of CN110310184A publication Critical patent/CN110310184A/en
Application granted granted Critical
Publication of CN110310184B publication Critical patent/CN110310184B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/12—Computing arrangements based on biological models using genetic models
    • G06N3/126—Evolutionary algorithms, e.g. genetic algorithms or genetic programming
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q30/00—Commerce
    • G06Q30/06—Buying, selling or leasing transactions
    • G06Q30/0601—Electronic shopping [e-shopping]
    • G06Q30/0611—Request for offers or quotes
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q30/00—Commerce
    • G06Q30/06—Buying, selling or leasing transactions
    • G06Q30/08—Auctions
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q50/00—Systems or methods specially adapted for specific business sectors, e.g. utilities or tourism
    • G06Q50/06—Electricity, gas or water supply

Abstract

The invention discloses a bidding simulation method for a power generation market based on a multi-agent game with a memory function, belonging to the technical field of power generation market simulation. Each generator looks as an agent, and then the bidding behavior among the generators is regarded as a multi-agent game problem to realize modeling and simulation; the power generation market is regarded as a complex self-adaptive system, and a main body in the market bidding process is packaged by a multi-agent technology; the generator intelligent agent has the function of memorizing the previous game strategy, and the model comprises an objective function, a constraint condition, a profit calculation module and a strategy adjustment algorithm module; and calculating bidding results according to market demands and bidding strategies of various power generators. The invention can solve the problem that the market balance is difficult to achieve in multi-agent power generation market bidding simulation; the system can assist power generator bidding personnel to make scientific decisions and assist a market supervision mechanism to conduct macroscopic regulation and control on the operation of a power generation market.

Description

Power generation market bidding simulation method based on multi-agent game with memory function
Technical Field
The invention belongs to the technical field of power generation market simulation, and particularly relates to a power generation market bidding simulation method based on a multi-agent game with a memory function.
Background
For a long time, the global power industry is operated in a traditional mode to implement a vertical integrated structure of power generation, transmission and distribution, but the defects of the mode are increasingly obvious.
After 'separation of plant networks and competitive bidding internet surfing' is carried out, a generator is used as an independent economic entity in a power generation market and needs to participate in competitive bidding internet surfing, and the generator needs to integrate information of the power generation market and a competitor and adopts different competitive bidding strategies to maximize profits of the generator. By regarding each power generator as an intelligent agent for realizing profit maximization by self and regarding bidding behaviors among the power generators as a modeling simulation method of a multi-intelligent-agent game problem, interaction characteristics and rules among the power generators and influence characteristics and rules of different bidding mechanisms and market rules on power generator interaction can be deeply researched. Decision support can be provided for bidding decision makers and also for the macroscopic monitoring of market regulators.
Disclosure of Invention
The invention aims to provide a power generation market bidding simulation method based on a multi-agent game with a memory function, which is characterized by comprising the following steps of: each generator looks as an agent, and then the bidding behavior among the generators is regarded as a multi-agent game problem to realize modeling and simulation; the power generation market is regarded as a complex self-adaptive system, and a main body in the market bidding process is packaged by a multi-agent technology; the generator intelligent agent has the function of memorizing the previous game strategy, and the model comprises an objective function, a constraint condition, a profit calculation module and a strategy adjustment algorithm module; the ISO agent can adopt an MCP or PAB bidding mechanism, and calculates bidding results according to market demands and bidding strategies of various power generators. The method is used for assisting a power generator bidding worker in making scientific decisions and assisting a market supervision mechanism in macroscopically regulating and controlling the operation of a power generation market; the method comprises the following specific steps:
step 1, in an initial stage, establishing a model for each generator agent, and acquiring own attributes of the generator agent by loading external data of the generator agent, wherein the data comprises the serial number, rated power, maximum/minimum power, marginal cost, post price, fixed cost data, variable cost data and environmental cost data of the generator agent; in the initial stage, the generator agent finishes self initialization by reading data from the outside; the initialized agent has the functions of self-adaptation and self-decision; independent System Operators (ISO) are considered a special class of agents, in addition to generator agents; the attributes of the ISO agent comprise market demand and bidding mechanism, and the behavior of the ISO agent is responsible for clearing the market;
step 2, after the initialization of the generator agent is completed, starting to enter a bidding state; the method comprises the following steps that the intelligent bodies of the power generators randomly generate initial quotations and quotations according to the constraint ranges of the quotations and the quotations of the intelligent bodies of the power generators, all the intelligent bodies of the power generators submit the initial quotations and the quotations to the ISO intelligent bodies, and after the quotations of all the intelligent bodies of the power generators are generated, the system submits all the quotations and the quotations to the ISO intelligent bodies; after the ISO intelligent agents complete calculation, returning bidding results to each intelligent agent of the power generator; there are two bidding mechanisms in ISO, the MCP (mark clearing price) mechanism and the PAB (pay as bid) mechanism:
step 3, the generator intelligent bodies enter an optimization stage according to the received bidding results and the quotation and report information of all the generators in the previous round, and each generator intelligent body searches an optimal strategy by adopting an improved genetic algorithm according to the quotation results of the generators in the previous round of the other N-1 generator intelligent bodies; making adjustments to the decision of the user, namely adjusting the quotation and the report amount;
and 4, under an MCP mechanism, according to quotes and reports submitted by each generator, the ISO ranks the generators from low to high according to the quotes, and accumulates corresponding declared capacities, the quote of the last generator meeting the market demand is a market clearing price, the generators with the quotes lower than the market clearing price bid successfully, the generators with the quotes higher than the market clearing price bid fail to bid, and finally all generators successfully bidding are settled according to the clearing price.
The specific steps of finding the optimal by adopting the improved genetic algorithm in the step 3 comprise:
step 31: the objective function according to each generator agent can be described as Li＝max piWithout loss of generality, the objective function abstraction of agent optimization is described as the following mathematical expression:
Figure GDA0003573230300000021
wherein, the first and the second end of the pipe are connected with each other,
Figure GDA0003573230300000022
for decision variables, f (X) denotes the objective function, xi,min≤xi≤xi,maxRepresenting constraint conditions, and finding a group of feasible solutions meeting the constraint conditions by means of an improved genetic algorithm to enable the value of the objective function to reach the maximum value;
step 32: the optimization problem described in step 31 is globally searched using a modified genetic algorithm:
step 2a, generating an initial population X consisting of NP intelligent agents in an n-dimensional search space allowed by a decision variable X, wherein NP is the number of the intelligent agents in the population; within the entire population X, each agent Xi＝(xi,1,xi,2,…,xi,n) All represent a feasible solution to the problem to be optimized;
and 2b, constructing a fitness function according to the target function and the specific constraint condition, wherein the adopted construction method comprises the following steps: the generator aims at maximizing profit, which is the difference between the electricity selling income and the cost; this objective function has no direct analytical expression because it depends not only on its own bidding strategy but also on the competitors' bidding strategy; the strategy optimizing space of the power generator is strictly limited in the constraint range; because the improved genetic algorithm provided by the invention has global convergence and no non-negative requirement on the target function, the fitness function is directly equal to the target function;
step 2c, for the given optimization problem, adopting a universal fitness transformation method to obtain the fitness values of all individuals in the population, and obtaining the population based on the obtained fitness values
Figure GDA0003573230300000035
Each agent in the population is evaluated and sorted according to the fitness, wherein t is the evolution generation time, and NP is the number of agents in the population; according to the sorting result obtained by the t generation population according to the descending sorting mode, the strategy of selecting an operator is as follows: NP x r after ranking is directly eliminated and replaced by NP x r intelligent agents before ranking, and the replaced population is sorted again according to the value of fitness in a descending order, wherein r is more than 0 and less than 1;
step 2d, crossing vector cross recombination strategy, comprising:
(a) dividing the sorted population into A and B parts:
Figure GDA0003573230300000031
Figure GDA0003573230300000032
(b) carrying out hybridization pairing on the agents in the A and B groups according to a preset rule:
Figure GDA0003573230300000033
step 2e, the well-matched population has NP/2 groups, and for each group of agents, a random number theta is generatedc＝rand([0，1]) If thetac＜pcThen the directional cross-recombination rule is completed according to the following formula:
for agent implementations in group a:
Figure GDA0003573230300000034
for agent implementations in group B:
Figure GDA0003573230300000041
otherwise, performing mutation operations on the agents in each group, including:
(a) to each pairing group
Figure GDA0003573230300000042
Xi ofcjCalculated from the following formula:
Figure GDA0003573230300000043
(b)Difor the optimization direction information obtained based on the hybridization pairing group, the following rules are calculated:
Figure GDA0003573230300000044
in the formula dq＝rand([0,1]) (9)
Step 33 mutual exclusion relationship: when theta is satisfiedc＜PcAnd are paired
Figure GDA0003573230300000045
And (3) only performing cross recombination operation when the space points corresponding to the two agents are poor in fitness algorithm, otherwise performing mutation operation: the method comprises the following steps:
(a) for agents in group a:
Figure GDA0003573230300000046
(b) implementing the following intelligent agents in the group B:
Figure GDA0003573230300000047
wherein: ximThe calculation rule of (1) is:
Figure GDA0003573230300000048
t is the maximum evolution algebra.
θjIs one and xiVectors of the same dimension, each element in the vector being between (-theta)i,q,θi.q) And thetai,q＝rand([0,1])；
In order to ensure the convergence of the algorithm, the algorithm adopts a population elite retention strategy; for population size invariance in the t-th generation population, all phi are subjected to(t)And by cross-recombination and mutation operations
Figure GDA0003573230300000049
Evaluating according to the fitness of the intelligent agents from big to small, and preferentially selecting the top M intelligent agents as the current evolution result;
step 34: according to the genetic algorithm optimizing result in the step 3, each power generator agent obtains the candidate strategies of the current turn, at the moment, the agent executes an internal strategy adjusting module, and the specific steps are as follows:
(a) and all the power generator intelligent agents perform pre-clearing operation according to the obtained candidate item strategies: submitting the obtained candidate strategies to an ISO agent, calculating a bidding result by the ISO according to a specific MCP or PAB bidding mechanism, and returning the result to each generator agent;
(b) each generator agent calculates profits under the candidate strategies according to information returned by the ISO, if the profits are larger than the profits of the previous round, the strategy is selected to be updated, and if not, the strategy of the previous round is kept unchanged;
(C) and (3) when the iteration times are less than the preset times, returning to the step (2) to continue calculation, and otherwise, outputting a result.
The method has the advantages that the improved genetic algorithm based on cross recombination and mutation recombination mutual exclusion of the hybrid vectors is adopted, the global optimizing capability of the intelligent agent is improved, and the decision robustness of the intelligent agent is enhanced. The simulation method can assist bidding decision-making personnel in making scientific decisions and can assist a market supervision mechanism in macroscopically monitoring the operation of the power generation market.
Drawings
FIG. 1 illustrates a MCP bidding mechanism.
FIG. 2 is a schematic diagram of a PAB bidding mechanism.
Fig. 3 is a cost composition diagram of a thermal power plant.
FIG. 4 is a schematic diagram of population replacement.
FIG. 5 is a schematic diagram of population pairing.
Figure 6 is a flow chart of the optimization algorithm.
FIG. 7 is a diagram of the simulation result of the quoted price of the power-generator under the MCP mechanism.
FIG. 8 is a diagram of the report simulation result of the power supplier under the MCP mechanism.
FIG. 9 is a diagram of a simulation result of market clearing under MCP mechanism.
Fig. 10 is a diagram of the simulation result of the quoted price of the power generator under the PAB mechanism.
FIG. 11 is a diagram of the telegraph volume simulation result under the PAB mechanism.
FIG. 12 is a graph of the simulation results of the market average bid price under the PAB mechanism.
Detailed Description
The invention provides a bidding simulation method of a power generation market based on a multi-agent game with a memory function.
The invention is further illustrated by the following figures and examples.
Firstly, each generator looks as an intelligent agent, and then the bidding behavior among the generators is regarded as a modeling of a multi-intelligent-agent game for simulation, and a memory function module is designed for the intelligent agent; the power generation market is regarded as a complex self-adaptive system, and a main body in the market bidding process is packaged by a multi-agent technology; the intelligent generator model comprises an objective function, a constraint condition, a profit calculation module and a strategy adjustment algorithm module, wherein the ISO intelligent module has the main function of calculating the market clearing price and the generating capacity of each generator according to market demand information and quotations and reports submitted by all generators; in order to improve the global optimizing capability of the intelligent agent, an improved genetic algorithm based on cross recombination and mutation recombination mutual exclusion of hybrid vectors is adopted; the method comprises the following specific steps:
step 1, in an initial stage, establishing a model for each power generator agent, and acquiring the attribute of the power generator agent by loading external data of the power generator agent, wherein the data comprises the serial number, rated power, maximum/minimum power, marginal cost, post price, fixed cost data, variable cost data and environment cost data of the power generator agent (as shown in a cost composition diagram of a thermal power plant shown in fig. 3); in the initial phase, the generator agent completes its initialization by reading data from the outside. The initialized agent has the functions of self-adaptation and self-decision. Independent System Operators (ISO) are considered a special class of agents, in addition to generator agents; the properties of the ISO agent include market demand and bidding mechanisms, and its behavior is responsible for clearing the market.
Step 2, after the initialization of the generator agent is completed, starting to enter a bidding state; the method comprises the following steps that the intelligent bodies of the power generators randomly generate initial quotations and quotations according to the constraint ranges of the quotations and the quotations of the intelligent bodies of the power generators, all the intelligent bodies of the power generators submit the initial quotations and the quotations to the ISO intelligent bodies, and after the quotations of all the intelligent bodies of the power generators are generated, the system submits all the quotations and the quotations to the ISO intelligent bodies; after the ISO intelligent agent completes calculation, returning bidding results to each generator intelligent agent; there are two bidding mechanisms in ISO, namely MCP (market clearing price) mechanism (shown in fig. 1 as MCP bidding mechanism diagram) and PAB (pay as bid) mechanism (shown in fig. 2 as PAB bidding mechanism diagram).
Step 3, the generator intelligent bodies enter an optimization stage according to the received bidding results and the quotation and report information of all the generators in the previous round, and each generator intelligent body searches an optimal strategy by adopting an improved genetic algorithm according to the quotation results of the generators in the previous round of the other N-1 generator intelligent bodies; making adjustments to the decision of the user, namely adjusting the quotation and the report;
and 4, under an MCP mechanism, according to quotes and reports submitted by each generator, the ISO ranks the generators from low to high according to the quotes, and accumulates corresponding declared capacities, the quote of the last generator meeting the market demand is a market clearing price, the generators with the quotes lower than the market clearing price bid successfully, the generators with the quotes higher than the market clearing price bid fail to bid, and finally all generators successfully bidding are settled according to the clearing price.
The specific steps of finding the optimum by using the improved genetic algorithm in step 3 (the optimization algorithm flowchart shown in fig. 6) include:
step 31: according to each hairThe objective function of an e-commerce agent may be described as Li＝max piWithout loss of generality, the objective function abstraction of agent optimization is described as the following mathematical expression:
Figure GDA0003573230300000071
wherein the content of the first and second substances,
Figure GDA0003573230300000072
for decision variables, f (X) denotes the objective function, xi,min≤xi≤xi,maxRepresenting constraint conditions, and finding a group of feasible solutions meeting the constraint conditions by means of an improved genetic algorithm to enable the value of the objective function to reach the maximum value;
step 32: the optimization problem described in step 31 is globally searched using a modified genetic algorithm:
step 2a, generating an initial population X consisting of NP intelligent agents in an n-dimensional search space allowed by a decision variable X, wherein NP is the number of the intelligent agents in the population; within the entire population X, each agent Xi＝(xi,1,xi,2,…,xi,n) All represent a feasible solution to the problem to be optimized;
and 2b, constructing a fitness function according to the target function and the specific constraint condition, wherein the adopted construction method comprises the following steps: the generator aims at maximizing profit, which is the difference between the electricity selling income and the cost; this objective function has no direct analytical expression because it depends not only on its own bidding strategy but also on the competitors' bidding strategy; the strategy optimizing space of the power generator is strictly limited in the constraint range; the improved genetic algorithm provided by the invention has global convergence and no non-negative requirement on the objective function, and the fitness function is directly equivalent to the objective function.
Step 2c, for the given optimization problem, adopting a universal fitness transformation method to obtain the fitness values of all individuals in the population, and obtaining the population based on the obtained fitness values
Figure GDA0003573230300000073
Each agent in the population is evaluated and sorted according to the fitness, wherein t is the evolution generation time, and NP is the number of agents in the population; according to the sorting result obtained by the t generation population according to the descending sorting mode, the strategy of selecting an operator is as follows: NP x r after ranking is directly eliminated and replaced by NP x r intelligent agents before ranking, and the replaced population is sorted again according to the value of fitness in a descending order, wherein r is more than 0 and less than 1; (the population replacement schematic shown in FIG. 4);
step 2d, crossing vector cross recombination strategy, comprising:
(a) dividing the sorted population into A and B parts:
Figure GDA0003573230300000074
Figure GDA0003573230300000081
(b) carrying out hybridization pairing on the agents in the A and B groups according to a preset rule: (as shown in the schematic diagram of population pairing in FIG. 5),
Figure GDA0003573230300000082
step 2e, the well-matched population has NP/2 groups, and for each group of agents, a random number theta is generatedc＝rand([0，1]) If thetac＜pcThen the directional cross-recombination rule is completed according to the following formula:
for agent implementations in group a:
Figure GDA0003573230300000083
for agent implementations in group B:
Figure GDA0003573230300000084
otherwise, performing mutation operations on the agents in each group, including:
(a) to each pairing group
Figure GDA0003573230300000085
Xi ofc,jCalculated from the following formula:
Figure GDA0003573230300000086
(b)Difor the optimization direction information obtained based on the hybridization pairing group, the following rules are calculated:
Figure GDA0003573230300000087
in the formula dq＝rand([0,1]) (9)
Step 33 mutual exclusion relationship: when theta is satisfiedc＜PcAnd are paired
Figure GDA0003573230300000088
And (3) only performing cross recombination operation when the space points corresponding to the two agents are poor in fitness algorithm, otherwise performing mutation operation: the method comprises the following steps:
(a) for agents in group a:
Figure GDA0003573230300000089
(b) for agents in group B:
Figure GDA00035732303000000810
wherein: ximThe calculation rule of (1) is:
Figure GDA00035732303000000811
t is the maximum evolution algebra.
θjIs one and xiVectors of the same dimension, each element in the vector being between (-theta)i,q,θi.q) And thetai,q＝rand([0,1])
In order to ensure the convergence of the algorithm, the algorithm adopts a population elite retention strategy; for population size invariance in the t-th generation population, all phi are treated(t)And by cross-recombination and mutation operations
Figure GDA0003573230300000092
Evaluating according to the fitness of the intelligent agents from big to small, and preferentially selecting M intelligent agents in the top as the result of the current evolution;
step 34: according to the genetic algorithm optimizing result in the step 3, each power generator agent obtains the candidate strategies of the current turn, at the moment, the agent executes an internal strategy adjusting module, and the specific steps are as follows:
(a) and all the power generator intelligent agents perform pre-clearing operation according to the obtained candidate item strategies: submitting the obtained candidate strategies to an ISO agent, calculating a bidding result by the ISO according to a specific bidding mechanism (MCP or PAB), and returning the result to each generator agent;
(b) each generator agent calculates profits under the candidate strategies according to information returned by the ISO, if the profits are larger than the profits of the previous round, the strategy is selected to be updated, and if not, the strategy of the previous round is kept unchanged;
(C) and (3) when the iteration times are less than the preset times, returning to the step (2) to continue calculation, and otherwise, outputting the result.
Examples
Five power generators in a certain area are selected as an example for analysis, and the relevant data of the five power generators are shown in tables 1 and 2:
TABLE 1 Generator basic data
Figure GDA0003573230300000091
TABLE 2 Power Generator cost data
Figure GDA0003573230300000101
Load demand aspect:
a) assuming that under the MCP mechanism, the market demand is 2500MW, the result of the competitive game of the power generator (shown in FIGS. 7, 8 and 9) is as follows:
b) assuming that under the PAB mechanism, the market demand is 2500MW, the result of the competitive game of the power generator (shown in figures 10, 11 and 12) is as follows:
as can be seen from fig. 7-9, under the MCP bidding mechanism, the generator achieves market equilibrium after about 400 rounds of gaming, and the uniform clearing price under the market equilibrium is about 316 yuan/MWh; as can be seen from fig. 10 to 12, under the PAB bidding mechanism, the generator achieves market equilibrium after about 100 rounds of game play, and the average clearing price under the market equilibrium is about 397 yuan/MWh; comparing fig. 9 and fig. 12, it can be seen that the advantage of MCP mechanism is far superior to PAB mechanism for the overall welfare of society in case of 89% of total supply (2800MW) of market demand (2500 MW).
The bidding decision staff of the power generator can obtain the bidding result under the balanced state of the market by setting the parameters (including the power generation capacity and the cost data of all the power generators) and the bidding mechanism of the power generator in the market, and the prediction result has guiding effect on the bidding strategy of the power generator. The market monitoring mechanism can predict market equilibrium results under various supply and demand conditions, various generator compositions, various macro regulation measures and different bidding mechanism conditions through model simulation, and further scientifically select a proper bidding mechanism and formulate reasonable and feasible macro regulation measures.

Claims (1)

1. A power generation market bidding simulation method based on a multi-agent game with a memory function is characterized in that each power generator is regarded as an agent, bidding behaviors among the power generators are regarded as modeling of the multi-agent game for simulation, and a memory function module is designed for the agent; the power generation market is regarded as a complex self-adaptive system, and a main body in the market bidding process is packaged by a multi-agent technology; the intelligent generator model comprises an objective function, a constraint condition, a profit calculation module and a strategy adjustment algorithm module, wherein the ISO intelligent module has the function of calculating the market clearing price and the generating capacity bid by each generator according to the market demand information and the quotations and the report submitted by all generators; in order to improve the global optimizing capability of the intelligent agent, an improved genetic algorithm based on cross recombination and mutation recombination mutual exclusion of hybrid vectors is adopted; the method specifically comprises the following steps: step 1, initializing a generator agent, establishing a model for each generator agent in an initial stage, and loading external data by the generator agent to acquire own attributes, wherein the data comprises the serial number, rated power, maximum/minimum power, marginal cost, benchmarking electricity price, fixed cost data, variable cost data and environmental cost data of the generator; in the initial stage, the generator agent finishes self initialization by reading data from the outside; the initialized intelligent agent has the functions of self-adaptation and self-decision; apart from the generator agents, the independent system operators ISO are regarded as a class of special agents; the attributes of the ISO agent comprise market demand and bidding mechanism, and the behavior of the ISO agent is responsible for clearing the market;
step 2, after the initialization of the generator agent is completed, starting to enter a bidding state; the method comprises the following steps that the intelligent bodies of the power generators randomly generate initial quotations and quotations according to the constraint ranges of the quotations and the quotations of the intelligent bodies of the power generators, all the intelligent bodies of the power generators submit the initial quotations and the quotations to the ISO intelligent bodies, and after the quotations of all the intelligent bodies of the power generators are generated, the system submits all the quotations and the quotations to the ISO intelligent bodies; after the ISO intelligent agent completes calculation, returning bidding results to each generator intelligent agent; there are two bidding mechanisms in ISO, MCP and PAB mechanisms:
step 3, the generator intelligent bodies enter an optimization stage according to the received bidding results and the quotation and report information of all the generator intelligent bodies in the previous round, and each generator intelligent body searches an optimal strategy by adopting an improved genetic algorithm according to the quotation results of the previous round of the other N-1 generator intelligent bodies; making adjustments to the decision of the user, namely adjusting the quotation and the report amount;
step 4, under an MCP mechanism, according to quotes and reports submitted by each generator, the ISO sorts the generators from low to high according to the quotes, accumulates corresponding declared capacities, takes the quote of the last generator meeting the market demand as a market clearing price, successfully bids by generators with quotes lower than the market clearing price, fails to bid by generators with quotes higher than the market clearing price, and finally settles accounts according to the clearing price uniformly by all generators with successful bids;
the specific steps of finding the optimal by adopting the improved genetic algorithm in the step 3 comprise:
step 31: the objective function abstraction of agent optimization is described as the following mathematical expression:
Figure FDA0003573230290000011
wherein the content of the first and second substances,
Figure FDA0003573230290000012
for decision variables, f (X) denotes the objective function, xi,min≤xi≤xi,maxRepresenting constraint conditions, and finding a group of feasible solutions meeting the constraint conditions by means of an improved genetic algorithm to enable the value of the objective function to reach the maximum value;
step 32: the optimization problem described in step 31 is globally searched using a modified genetic algorithm:
step 2a, generating an initial population X consisting of NP intelligent agents in an n-dimensional search space allowed by a decision variable X, wherein NP is the number of the intelligent agents in the population; within the entire population X, each agent Xi＝(xi,1,xi,2,…,xi,n) All represent a feasible solution to the problem to be optimized;
and 2b, constructing a fitness function according to the target function and the specific constraint condition, wherein the adopted construction method comprises the following steps: the generator aims at maximizing profit, which is the difference between the electricity selling income and the cost; this objective function has no direct analytical expression because it not only depends on its own bidding strategy, but also is related to the competitor's bidding strategy; the strategy optimizing space of the power generator is strictly limited in the constraint range; because the improved genetic algorithm has global convergence and no non-negative requirement on the objective function, the fitness function is directly equal to the objective function;
step 2c, for the given optimization problem, adopting a fitness transformation method to obtain the fitness values of all individuals in the population, and obtaining the population based on the obtained fitness values
Figure FDA0003573230290000021
Each agent in the system is evaluated and ranked according to the fitness of the agent, and t is the evolution generation time; according to the sorting result obtained by the t generation population according to the descending sorting mode, the strategy of selecting an operator is as follows: NP x r intelligent agents after ranking are directly eliminated and replaced by NP x r intelligent agents before ranking, and the population after replacement is sorted again according to the value of fitness, wherein 0<r<1；
Step 2d, crossing vector cross recombination strategy, comprising:
(a) dividing the sorted population into A and B parts:
Figure FDA0003573230290000022
Figure FDA0003573230290000023
(b) carrying out hybridization pairing on the agents in the A and B groups according to a preset rule:
Figure FDA0003573230290000024
step 2e, the matched population has NP/2 groups, and for each group of agents, a random number theta is generatedc＝rand([0,1]) If θ c<PcThen the directional cross recombination is completed according to the following formula:
for agent implementations in group a:
Figure FDA0003573230290000025
for agent implementations in group B:
Figure FDA0003573230290000026
otherwise, performing mutation operation on the agents in each group;
(a) xi of each pairing groupc,jCalculated from the following formula:
Figure FDA0003573230290000031
(b)Difor the optimization direction information obtained based on the hybridization pairing group, the following rules are calculated:
Figure FDA0003573230290000032
in the formula dq＝rand([0,1]) (9)，
Step 33: mutual exclusion relationship: when theta is satisfiedc<PcAnd are paired
Figure FDA0003573230290000033
And (3) only performing cross recombination operation when the space points and fitness algorithms corresponding to the two intelligent agents are different, otherwise performing mutation operation: the method comprises the following steps:
(a) for agents in group a:
Figure FDA0003573230290000034
(b) for agents in group B:
Figure FDA0003573230290000035
wherein: ximThe calculation rule of (1) is:
Figure FDA0003573230290000036
t is the maximum evolution algebra;
θjis a and XiVectors of the same dimension, each element in the vector being between (-theta)i,q,θi.q) And thetai,q＝rand([0,1])；
Adopting a population elite retention strategy; for population size invariance in the t-th generation population, all phi are treated(t)And by cross-recombination and mutation operations
Figure FDA0003573230290000037
Evaluating according to the fitness of the intelligent agents from big to small, and preferentially selecting M intelligent agents in the top as the result of the current evolution;
step 34: according to the genetic algorithm optimizing result in the step 33, each power generator agent obtains the candidate strategy of the current turn, and at the moment, the agent executes an internal strategy adjusting module, and the specific steps are as follows:
(a) and all the power generator intelligent agents perform pre-clearing operation according to the obtained candidate item strategies: submitting the obtained candidate strategies to an ISO intelligent agent, calculating a bidding result according to a specific bidding mechanism by the ISO, and returning the result to each generator intelligent agent;
(b) each generator agent calculates profits under the candidate strategies according to information returned by the ISO, if the profits are larger than the profits of the previous round, the strategy is selected to be updated, and if not, the strategy of the previous round is kept unchanged;
(C) and (3) when the iteration times are less than the preset times, returning to the step (2) to continue calculation, and otherwise, outputting the result.
CN201910615663.7A 2019-07-09 2019-07-09 Power generation market bidding simulation method based on multi-agent game with memory function Active CN110310184B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201910615663.7A CN110310184B (en) 2019-07-09 2019-07-09 Power generation market bidding simulation method based on multi-agent game with memory function

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201910615663.7A CN110310184B (en) 2019-07-09 2019-07-09 Power generation market bidding simulation method based on multi-agent game with memory function

Publications (2)

Publication Number Publication Date
CN110310184A CN110310184A (en) 2019-10-08
CN110310184B true CN110310184B (en) 2022-05-10

Family

ID=68079921

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201910615663.7A Active CN110310184B (en) 2019-07-09 2019-07-09 Power generation market bidding simulation method based on multi-agent game with memory function

Country Status (1)

Country Link
CN (1) CN110310184B (en)

Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107845022A (en) * 2017-11-02 2018-03-27 北京恒泰能联科技发展有限公司 Electricity market aid decision-making systems
CN108985897A (en) * 2018-07-12 2018-12-11 华东交通大学 A kind of smart grid Generation Side Differential evolution game price competing method

Family Cites Families (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20140006171A1 (en) * 2012-06-29 2014-01-02 Yahoo! Inc. Method of offline experimentation environment and apparatus conducting the same

Patent Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107845022A (en) * 2017-11-02 2018-03-27 北京恒泰能联科技发展有限公司 Electricity market aid decision-making systems
CN108985897A (en) * 2018-07-12 2018-12-11 华东交通大学 A kind of smart grid Generation Side Differential evolution game price competing method

Non-Patent Citations (3)

* Cited by examiner, † Cited by third party
Title
A new enhanced bat-inspired algorithm for finding linear supply function equilibrium of GENCOs in the competitive electricity market;Taher Niknam等;《Energy Conversion and Management》;20131231;第76卷;第1015-1028页 *
一种电力市场环境下的电源规划多智能体模型;黄仙 等;《电力系统保护与控制》;20161216;第44卷(第24期);第1-8页 *
基于复杂自适应系统架构的发电商竞价仿真研究;张军;《中国优秀博硕士学位论文全文数据库（硕士） 工程科技Ⅱ辑》;20210615(第6期);第C042-213页 *

Also Published As

Publication number Publication date
CN110310184A (en) 2019-10-08

Similar Documents

Publication Publication Date Title
Mei et al. An efficient feature selection algorithm for evolving job shop scheduling rules with genetic programming
Wen et al. Optimal bidding strategies and modeling of imperfect information among competitive generators
Shao et al. Multi-objective evolutionary algorithm based on multiple neighborhoods local search for multi-objective distributed hybrid flow shop scheduling problem
Sensfuß et al. Agent-based simulation of electricity markets. A literature review
Son et al. Hybrid coevolutionary programming for Nash equilibrium search in games with local optima
Deb Multi-objective optimisation using evolutionary algorithms: an introduction
Mendoza et al. NSGA and SPEA applied to multiobjective design of power distribution systems
Martinez-Jaramillo et al. An heterogeneous, endogenous and coevolutionary GP-based financial market
Nikoofard et al. Multiobjective invasive weed optimization: Application to analysis of Pareto improvement models in electricity markets
Zaman et al. Co-evolutionary approach for strategic bidding in competitive electricity markets
CN110109753A (en) Resource regulating method and system based on various dimensions constraint genetic algorithm
Lorenz et al. Optimal algorithms for k-search with application in option pricing
Shim et al. A hybrid estimation of distribution algorithm for solving the multi-objective multiple traveling salesman problem
Wen et al. A genetic algorithm based method for bidding strategy coordination in energy and spinning reserve markets
Tsaousoglou et al. Flexibility aggregation of temporally coupled resources in real-time balancing markets using machine learning
CN110310184B (en) Power generation market bidding simulation method based on multi-agent game with memory function
Barkat Ullah et al. Search space reduction technique for constrained optimization with tiny feasible space
CN111784440A (en) Game electricity purchasing bidding method and system, terminal device and storage medium
CN115146455B (en) Complex supply chain multi-objective decision method supported by calculation experiment
Zaman et al. A co-evolutionary approach for optimal bidding strategy of multiple electricity suppliers
He et al. EABOA: Enhanced adaptive butterfly optimization algorithm for numerical optimization and engineering design problems
CN111798024A (en) Method and device for determining power plant quotation under subsection quotation rule
Bajpai et al. Strategic bidding in network constrained electricity markets using FAPSO
Richter et al. Effects of tree size and state number on gp-automata bidding strategies
Esen Utilizing genetic algorithm to detect collusive opportunities in deregulated energy markets

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant