Research On Electric Taxi Operation Simulation Based On Multi-agent Technology And Reinforcement Learning Algorithm

Electric vehicles connected to the electricity grid will have a great impact on the power system,especially with the high penetration of electric vehicles,the safety and stability of the power system could be threatened.Generally,Electric taxi operation has three characteristics: long operation time,long distance and high charging density.And its operating behavior is affected by the distribution of passengers,traffic conditions,charging price and other factors,showing a greater random.Therefore,the uncertainty to power system caused by the power grid will also be greater.It is difficult for Monte Carlo method and traditional mathematical optimization to establish a certain mathematical model between main body and the environment.In order to accurately simulate the electric taxi operation behavior and load characteristic,this paper proposes a mothed based on multi-agent technique and reinforcement learning algorithm to study the operation behavior of electric taxis.Two kinds of reinforcement learning algorithms are compared to study the influence between electric taxi operating behavior and charging load.On this basis,the paper discusses the influences of shift mode and charging guidance on the operation of electric taxis.The main work of this paper is as follows:Firstly,the framework of electric taxi real time operation simulation system is constructed based on multi-agent technology.In the framework,this paper builds various types of agent model based on JADE development platform.The interaction mechanism of agents' information and interface file are designed.In order to achieve the distributed simulation of electric taxi operation,the agents are divided into a number of modules distributed in different hosts to run according to the characteristics of different types of agents.Secondly,two reinforcement learning algorithms are compared to study the influence on electric taxi operation behavior.A variety of learning algorithms are systematically introduced.Then,the reinforcement learning algorithm model of electric taxi agent is established from five aspects: state space,behavior decision space,behavior strategy selection and probability updating and reward and punishment function.A Q(?)learning algorithm with multi-step foresight capability is developed,and compared with the Q-Learning learning algorithm from different dimension.On this basis,the influence of SOC threshold and time distribution of shift mode on the operation and charging load of electric taxi is discussed.Thirdly,different charging strategies on the operation of electric taxi are studied.Based on the charging strategy of shortest path,a charging guidance strategy model is proposed,which considers the charging distance,the charging queue time and the charging equilibrium degree of the charging equipment.The simulation results show that the electric taxi adopting the charging guidance strategy will not only decrease the charging queue time,improve the operation time and income of the electric taxi,but also help improve the equilibrium degree and utilization rate of the charging equipment,reduce the power grid loss and voltage offset.