Optimization Strategy Of Distribution Network With Distributed Power Based On Deep Reinforcement Learning

With the development of distributed power technology,it is inevitable for wind,light and other new energy to be connected to the distribution network,which can effectively alleviate the environmental crisis.However,the randomness and volatility of distribution network control also bring new challenges.The development of energy storage technology promotes the absorption of new energy.The flexible charge-discharge characteristics of energy storage devices enable it to cooperate with new energy to control the distribution network.However,the access of a large number of Distributed Generation（DG）increases the complexity of the distribution network structure.Therefore,it has great research value and practical significance to improve the security and economy of distribution network operation through distributed power control.This thesis mainly studies the optimization strategy of distribution network with distributed power supply.The main work of this thesis is as follows:Firstly,this thesis introduces the wide application of distributed power generation technology in various countries and the research status of distribution network optimization with distributed power in domestic and foreign.Then,the structure of distribution network is simplified based on graph theory,introduce the power flow calculation method,and analyze the influence of distributed power grid connection on distribution network.Select the IEEE 33 node system,and calculate the power flow calculation by MATLAB,and analyze the influence of distributed power supply access location,and output on distribution network voltage and active power network loss.This part of the thesis propose a global optimization strategy of distribution network based on deep reinforcement learning,which can optimize the operation of distribution network by scheduling distributed power supply reasonably in real time according to given states.Taking the loss cost of distribution network as the optimization objective,considering the security constraints of distributed power supply and distribution network operation,a mathematical model was established,and the real-time operation optimization strategy was solved by DDPG,which proved that the strategy obtained by deep reinforcement learning could reduce the network loss by controlling the output of distributed power supply.This part of the thesis put forward a kind of power distribution network based on the multi-agent deep reinforcement learning partition optimization strategy,according to the electric distance using spectral clustering algorithm for distribution network partitioning,each subdomain modeling as an agent,and the center type training-the architecture of distributed execution,the intelligence that can be learned in the process of training coordination control strategy,based on the local observation information,real-time decision can be made to optimize the distribution network.Taking the distribution network voltage offset as the optimization objective,a mathematical model is constructed,and build Markov game,the framework of central type training-distributed execution and adopt the strategy of Twin Delayed Deep Deterministic policy gradient algorithm（TD3）to solve the real time operation optimization strategy,proved through the deep multi-agent reinforcement learning strategy can achieve the goal of reducing voltage deviation.