The Research On Cross-realm Energy-Computation Scheduling Mechanism For Electric Vehicles Via Deep Reinforcement Learning

To achieve the goal of "carbon peaking and carbon neutrality",promoting energy transformation,which promoting the development and promotion of electric vehicles(EVs).In addition to being a means of transportation,due to the energy attributes of EVs,disorderly charging of a large-scale integration into the power grid can have a negative impact on the safe and stable operation of the power grid.However,as highly flexible mobile loads and energy storage units,EVs can interact with the power grid through reasonable incentive mechanisms.In addition,the advancement of information and communication technology has promoted the intelligent development of EVs and charging stations(CSs),leading to new edge computing and intelligent charging network transactions.As an edge computing node,EVs can provide flexible computing offload services for CSs during the charging process.Therefore,the present study will investigate the cross-realm scheduling of EVs from the following three aspects.Firstly,due to the impact of traffic system when scheduling EVs to participate in ancillary service,it is necessary to reasonably select vehicles and plan routes.Therefore,this paper introduces real-time traffic information to construct a power-transportation interaction system to assist in peak shaving scheduling.In order to address system uncertainty,this paper proposes an EVs peak shaving scheduling strategy based on the Cooperative Multi-agent Deep Reinforcement Learning algorithm QMIXBid,which implements CSs collaboration to coordinate the allocation of EVs based on QMIX,and designs a bidding mechanism to avoid repeated selection of EVs by multiple CSs through state action value.Experimental results show that the proposed distributed EVs scheduling strategy outperforms the baseline and can reduce the cost of peak shaving scheduling.Secondly,in order to coordinate the computational resources and energy transactions of EVs,the interaction between EVs and CSs and the system uncertainty are incorporated into the joint optimization of edge computing scheduling and energy trading,a reinforcement contract design algorithm based on Parametric Deep Q Network(PA-DQN)is proposed,which uses hybrid action space to design contracts for CSs,matching computing tasks,charging resources and EVs.By incorporating incentive compatibility constraints,individual rationality constraints,and system constraints into contract design,the participation of EVs is incentivized and long-term social welfare is maximized.In addition,the concept of social distance is introduced,and EVs decide whether to accept the unique contracts based on social distance.Experimental results show that the proposed algorithm outperforms the baseline and can approximate the optimal solution.Lastly,considering the intelligent transformation of the power grid,this paper proposes an EV-grid interaction platform based on deep reinforcement learning.With the support of the above research content,the platform adopts a front-end and back-end separation architecture using Django and Vue frameworks,and implements CSs and EVs monitoring with HELICS,and EV scheduling for peak shaving and EV edge computing scheduling and energy trading contract design based by deploying reinforcement learning models.