Microgrid Energy Management And Dispatching Based On Reinforcement Learning

The microgrid energy supply and demand system is characterized by intermittency and volatility,and there are also considerable differences in the forms of different energy storage technologies.The existence of these characteristics leads to the difficulties in fine control of the microgrid,and brings a series of challenges to the overall optimization of microgrid energy management and dispatching.In addition,the transmission line loss is significant due to the low-voltage characteristics of microgrid,but is often overlooked in the existing research.Therefore,based on the actual operating characteristics of microgrid,with economic operation and efficient dispatching as the goal,this dissertation will conduct an in-depth research on the energy management and dispatching approaches for microgrid.Considering the advantages of reinforcement learning in optimization and decision-making in complex,dynamic and uncertain environments,the study will be addressed based on the reinforcement learning approaches.The specific research issues include the development of the improved reinforcement learning algorithm,charging and discharging scheduling optimization for energy storage battery,storage configuration for integral energy,as well as energy demand response dispatching.The main work of the dissertation is summarized as follows.Firstly,a charging and discharging scheduling optimization model of microgrid energy storage battery is established and a cooresponding reinforcement learning optimization approach is proposed.In order to meet the need for lean management of energy storage battery charging and discharging,a scheduling optimization model for microgrid energy storage battery charging and discharging is built.For this model,an improved deep Q network algorithm with the sequence sample adaptive priority adjustment strategy is proposed to optimize the battery energy storage efficiency based on battery status,charging/discharging decisions,and electricity price.Experimental results demonstrate that the proposed approach can improve the operating efficiency of the microgrid energy storage system significantly.Secondly,an integral battery life loss based microgrid energy configuration optimization model is proposed.Improved lightweight neural network is presented for the quantitative evaluation of the battery life loss.Furthermore,combined with comprehensive energy supply and demand,a two-layer optimization model is constructed for capacity configuration and dispatching operation.Experimental results show that the proposed model can effectively improve the performance of microgrid operation dispatching.Finally,a hierarchical deep Q network approach is proposed for microgrid energy management and dispatching.According to the load demand for electrical energy and line loss,a mixed optimization model containing discrete and continuous variables is built,with the transmission power and battery charging/discharging action of each transmission line as the decision variables.Coorespondingly,a hierarchical deep Q network(HDQN)is proposed for energy collaborative optimal dispatching to provide the battery charging and discharging decision of the upper network and the transmission line power allocation of the lower network.Experiment results show that the proposed approach can achieve excellent economic benefits.