基于安全强化学习的主动配电网有功-无功协调优化调度OA北大核心CSTPCD

A safe reinforcement learning method based on offline strategies is proposed.Through offline training of a large amount of historical operating data of the distribution network,it gets rid of the traditional optimization method.Dependence on complete and accurate models.First,combined with the distribution network parameter information,an active and reactive power optimization model based on the constrained Markov decision process(CMDP)was established;then,a new safety reinforcement learning method was designed based on the original dual optimization method.The cost function is minimized while maximizing future discount rewards;finally,simulations are performed on power distribution system.The simulation results show that the proposed method can online generate a dispatching strategy that satisfies complex constraints and has economic benefits based on real-time observation information of the distribution network.