Cooperative And Optimal Scheduling Strategy For Wind/Solar/Storage Station Cluster Based On Reinforcement Learning

China has vigorously promoted the construction of new energy stations in the past decade to reduce carbon emissions from electricity production.However,the increase in the number of new energy stations has increased the difficulty of real-time optimization and scheduling of the power system,and the adjustable resources of the stations cannot be fully utilized.The problem of insufficient real-time solving ability of traditional optimization scheduling algorithms is becoming increasingly prominent.How to solve the optimization scheduling model of renewable energy faster and better has become an urgent problem to be solved.In recent years,data-driven deep reinforcement learning methods have received widespread attention due to their ability to effectively solve high-dimensional nonlinear complex problems.This method can provide a solution for the real-time optimization scheduling problem of wind and solar storage station clusters.The research content of this article is as follows:(1)Real time energy storage cycle life assessment method.Based on the rain flow counting method,convert a single,variable power charge/discharge instruction into a fixed power charge/discharge instruction for multiple energy storage units.Based on experimental data,evaluate the energy storage life loss generated by fixed power charging/discharging.Calculate the average energy storage loss of all energy storage units and estimate life loss of energy storage.Compare the real-time energy storage life assessment method proposed in this article with the energy storage life loss assessment model of other optimization scheduling algorithms,and analyze the effectiveness of the real-time energy storage cycle life assessment method.(2)Optimization scheduling method for station clusters based on deep deterministic strategy gradient algorithm.Build a station group optimization scheduling model,based on the reinforcement learning model,convert the solving process of the station group optimization scheduling model into a Markov Decision Process,in order to facilitate the application of reinforcement learning algorithms.Based on the real-time optimization requirements of station cluster optimization,the Actor-Critic architecture is applied to achieve distributed optimization scheduling.(3)Example analysis of real-time optimization scheduling algorithm for station clusters.Based on the real-time optimization scheduling requirements of station clusters,six scenarios were designed to verify the policy learning effectiveness of the deep deterministic strategy gradient algorithm.Comparing the training iterations required for the real-time energy storage cycle life assessment method with the rain flow counting method applied in previous literature,it is demonstrated that the real-time evaluation algorithm can improve training efficiency.Analyze the example results of the real-time optimization scheduling algorithm for the station cluster,and prove that the algorithm can meet the optimization scheduling requirements.Compare the optimization scheduling results of reinforcement learning algorithms and solvers,and analyze the advantages of reinforcement learning algorithms in real-time optimization scheduling problems.