A Review of Research on the Application of Deep Reinforcement Learning in Optimization Dispatch of Power Grids With Distributed Flexible Resources
Since China proposed the carbon peak and carbon neutrality goals in 2020,distributed flexible resources such as rooftop photovoltaic,electric vehicles,and flexible energy storage have exhibited a trend of massive development,providing significant potential for the balance of the new type power systems.However,as the multiple uncertainties of massive flexible resources increase,the spatiotemporal decision variables is becoming more complex and high-dimensional,and the difficulty of accurate mechanism modeling has surged sharply,causing traditional optimization methods to encounter bottlenecks when solving the power grid optimization dispatch problems with large-scale,highly random,and cognitively difficult flexible resources.In recent years,as a new generation of machine learning paradigm,deep reinforcement learning has demonstrated the ability to cope with such challenges by learning optimal strategies through interaction with the environment when there is no detailed model parameters.In this regard,the paper provides a comprehensive review of research on optimization dispatch of power grids with distributed flexible resources.Specifically,it first analyzes the operational characteristics of resources,problem modeling,and solution strategies.Then it briefly outlines the principles and classification of the algorithms.Following this,it divides scenarios into demand-side user energy management,aggregated layer cluster coordinated response,and grid-side optimization operation control according to the different focuses of the dispatch problem,analyzing typical applications,solution processes,algorithm effectiveness.Subsequently,it summarizes the advantages and disadvantages of existing methods,and suggests improvements.Finally,it analyzes future research directions from the perspectives of constructing simulation environments,improving solving strategies,and enhancing agent performance.
distributed flexible resourcesoptimization dispatchdeep reinforcement learningdata-driven methodnew type power systems
国家科技期刊平台
NSTL
万方数据
