Research On Differentiated Multi-objective Bidding Strategy Of Electricity Retailers

With the deepening of reform on China’s electric power system and the opening of power retail market,a large number of the retailers participate in market.The complexity of the retailers’ bidding behavior has presented a series of new challenge to the formulation of electricity market rules,the design of trading models,and the management of participants in electricity market.There are various retailers in China.These retailers have different backgrounds,including the type of gird retailers with experience in purchasing and selling electricity,the type of power generation retailers with experience in bidding for the grid,and some independent retailers without experience in purchasing and selling electricity.Retailers with different backgrounds have different purposes when participating in market transactions,so the bidding strategies are also different.At the same time,the same retailer does not participate in market bidding for a single purpose,but needs to make decisions among multiple bidding purposes(such as profit,electricity,turnover).Therefore it is necessary to study the differentiated multi-objective bidding strategies of the retailers in electricity market.In this thesis,the simulation of agent technology is used to study the differentiated multi-objective bidding strategy of the retailers under relevant bidding rules of monthly-centralized electricity market.The relevant results can provide some reference for the simulation of the electricity market,the bidding of the ratailers and the formulation of market trading rules in the future.1.Based on the agent modeling technology,intelligent agent bidding models of the electricity market members are established.Firstly,the interaction among the market members and market transaction rules is analyzed in market simulation framework.The cost and benefit models of each market participants are built.Then,the Roth-Erev(RE)reinforcement learning algorithm is used to establish an intelligent agent decision model.The applicability of the reinforcement learning algorithm to simulate the bidding behavior of the retailers is verified,which laid the foundation for the simulation research in the subsequent chapters.2.Considering the differentiated factors of retailers,differentiated single-objective bidding agent model is established.The impact of the composition of assets,operation objectives,the share of electricity market,and the user contracts are considered in this model.The impact of different bidding objectives,user sharing models and self-learning behaviors on the bidding results is qualitatively compared.Studies show that the differentiated factors of the retailers will lead to different bidding objectives and differences in the distribution of benefits.The bidding strategies are also changed with different factors of the retailers.The formulation of market rules is provided by relevant simulations.3.The multi-objectives bidding agent model of the retailers is considered to establish based on operational objectives of the retailers such as power demand,profit and turnover.Under the two clearing mechanisms of Pay-as-Clear(PAC)and Pay-As-Bid(PAB),the changes of the retailers’ multiobjective bidding strategies are studied with weighting method and interactive decision method.Research shows that when retailers use multi-objective bidding strategy,the preference of the objective ratio will lead to changes in the convergence speed and convergence range.The interactive decision-making method can reflect the influence on the objective’s preference on the bidding strategy more than the weighting method in the process of deciding on multi-objective.The relevant simulation results provide some reference for the formulation of market rules.