Research On Intellectualization Of Game Behavior In Power Market

Under the new round of power reform policies in China,the power market has gradually opened up.At present,the power market in China is mainly composed of medium and longterm transactions and power spot markets.A market-oriented energy balance mechanism with medium and long-term transactions as the mainstay and spot transactions as a supplement is gradually established.The power grid usually contains multiple power generation companies,forming a competitive relationship.Power generation companies will maximize their own interests in the power market environment,which will form a game when selling power.However,each power generation company is not completely rational and has different information when making strategies,so there are risk of affecting market economy and stability.Based on the background of appeal,this paper will study the game behavior of power generation companies in the power market,which mainly includes two aspects:1.This dissertation studies the behavior of single game in the power market,and introduces an integrated forecasting model based on single game that includes machine learning models and power systems.This model uses photovoltaic(PV)power generation plant as an example to construct an integrated prediction model of PV power generation plant and power grids based on a single game.The model takes into account the benefits of both PV power generation plant and the grid,and aims to reduce industrial costs by using forecast errors rather than simply aiming at accuracy.The feedforward neural network is selected as a typical machine learning model for integration.In addition,in order to solve the nested optimization problem in network training,an equivalent derivable model is established to make the objective function derivable,so that the gradient-based optimization algorithm is used.The simulation results show that the integrated prediction model based on the single game proposed in this paper is more economical and effective than the traditional pure precision prediction model.It can improve the advantages and benefits of PV power generation plant in the market game,and increase the winning bid rate and scalar.2.Study the behavior of multi-body game in power market,and build a multi-body game model of power market.This paper uses multiple power generation plants as market players to build a multi-period non-cooperative intelligent multi-body game model.Each power generation plant in the model has autonomy and an independent agent,which can intelligently adjust the game strategy of the power generation plant based on the game results.The machine learning method used by the agent is Q-learning algorithm.This paper simulates and analyzes the multi-body non-cooperative game model through two market environments.The simulation results show that: 1)Power generation plants with lower costs have more advantages in the initial stage of the game,but after a long-term game,their advantages are replaced by higher-cost power generation plants;2)Compared with scenario 2,plant benefits in scenario 1 do not converge and LMP values rise significantly,so the market environment is at risk of collapse,which is not conducive to market economy and stability.For this market environment,certain market policies should be set to ensure the normal operation of the market.