Research On Optimal Scheduling Of Microgrid Based On Reinforcement Learning

Nowadays,the world is facing the crisis of imbalance between supply and demand of energy system due to the shortage of fossil resources and the rapid growth of power demand,as well as the ecological and environmental problems such as global warming and ozone layer destruction.Renewable energy with the characteristics of energy conservation,emission reduction and environmental friendliness has become the core of promoting the transformation and development of the world energy economy.As a new generation technology,microgrid can achieve the goals of success rate balance and system operation optimization with the control and energy management of the system itself.In order to fully exploit the advantages of microgrid and promote the use of renewable energy,it is of great theoretical and practical significance to study the optimal dispatch of microgrid.Aiming at the problem that the current microgrid scheduling optimization methods can not effectively accumulate and exploit the learned scheduling knowledge,this paper takes the typical microgrid system including wind solar distributed generation,diesel generator and battery as the research object,and introduces self-study and reinforcement learning of knowledge accumulation ability into the microgrid scheduling optimization problem.Firstly,this paper studies the reinforcement learning theory,summarizes the basic elements of reinforcement learning model,and the learning principles of different types of reinforcement learning algorithms.Secondly,the operation mechanism and related characteristics of typical micro grid system internal units are analyzed,and the corresponding mathematical model is constructed,as well as the operation cost objective function including diesel generator fuel cost and micro grid and large grid transaction cost.Then,the models of state,action space and reward function of microgrid optimal dispatch based on deep double-q network and deep deterministic strategy gradient are established.The effectiveness of the microgrid optimal dispatch model based on deep double-q network and the microgrid optimal dispatch model based on deep deterministic strategy gradient is verified by a specific example,Deep deterministic strategy gradient has more advantages in solving microgrid optimal scheduling problem,and can obtain better scheduling strategy.Then,considering the effective utilization of the accumulated scheduling knowledge,this paper introduces transfer learning and combines it with deep deterministic strategy gradient to propose microgrid optimal scheduling based on deep deterministic strategy gradient and transfer learning,The knowledge transfer rules based on the similarity of actual power demand and supply are designed,and then the proposed method is verified by a specific example.The results show that the design of knowledge transfer rules and the proposed method are effective in realizing the accumulation,mining and utilization of scheduling knowledge.