Research On EV Charging And Discharging Game Strategy Based On Deep Reinforcement Learning

Global energy supply and demand are tight,and countries around the world are striving to achieve "carbon neutralization".The development of energy conservation and new energy sources has become the mainstream trend.With the popularity of electric vehicle(EV)and its low pollution and noise-free characteristics,it has been developed rapidly all over the world.However,large user clusters tend to charge during the night rush hour,which will not only lead to a surge in night power demand,but also increase user charging costs.How to formulate a community EV charging strategy to maximize the balanced interests of both EV agents and EV users is an important part of power system research.Traditional power system optimization strategies often only take into account the single-party model on the power grid side or demand side,but ignore the game strategy model under the joint action of both sides,resulting in a loss of the optimization results.Based on the above reasons,this thesis establishes the charging and discharging game strategy model of EV cluster according to the needs of both user clusters and agents,and solves the optimal game solution under the specified conditions to achieve the goal of maximizing balanced interests.The following research work has been carried out:(1)The classical EV charging two-party model and charging model are modeled.The travel data of electric vehicles with Chinese user behavior characteristics are selected and fitted with functions.This thesis models the profit of both EV users and EV agents,simulates the model using Monte Carlo algorithm and PSO algorithm,studies the relationship between the benefits of both parties,and puts forward the problems of the classic model.(2)Electricity price data prediction model based on LSTM neural network.The features of electricity price data and the advantages of extracting electricity price features using LSTM neural network are introduced.The LSTM long-term and short-term memory network and ARIMA algorithm are used to predict historical electricity price data,and the advantages and disadvantages of the two algorithms are compared.(3)EV charging and discharging game strategy model based on MADDPG multi-agent algorithm.This thesis compares the multi-agent algorithms and combines the research features of the subject to select the appropriate algorithm,uses the multi-agent algorithm to build a Markov game,creates an action-state space function and an EV charging two-party model objective function,simulates the algorithm model to find the game balance solution,and compares the multi-value selection of the hyper-parameters.The charging and discharging game model is simulated by using multi-agent algorithm and electricity price prediction model.The algorithm model proposed in this thesis improves the charging satisfaction of users by 4.6% and 6.1% compared with DDPG algorithm and classic PSO algorithm,and improves the profit of agents by 4.1% and 6.7%,respectively.The validity of the method mentioned in this thesis is verified.