Topology Information Reconstruction And Optimal Regulation Technology Of Partially Visible Distribution Networks

In order to facilitate the implementation of China’s carbon neutrality strategy,more and more Distributed Generation(DG)are connected to the distribution network.But DG,with its randomness,intermittency and fluctuation,brings new problems and challenges to the optimization and regulation of distribution network.At the same time,due to the lag of the development of measuring devices and communication facilities in distribution network,the state of the feeder system interconnection switches and segmentation switches as well as the load data of some nodes in the station system can not be obtained in real time,resulting in the feeder system and the system in distribution network are in a state of partially visible.In order to realize optimal regulation of low perception distribution network,the following research works are carried out in this thesis focusing on topology information reconstruction and optimal regulation technology of distribution networks:1)Considering that the status of the interconnection switch and segmentation switch in the feeder system could not be obtained in real time,the traditional optimization method could not be used effectively.In this paper,the structure of the traditional graph convolutional Network was improved based on the Residual Network(Res Net)structure,and the reconstruction method of the topology information of the feeder system was studied.It lays the physical model foundation for feeder system optimization and regulation.Compared with the method based on the correlation analysis of long time series data,this method can quickly judge the status of the feeder system’s interconnection switch and segmentation switch according to the voltage data on a single time section.2)Considering the voltage fluctuation and voltage overlimit caused by the access of DG to distribution network,this paper carries out probabilistic modeling of DG and load,and optimizes and regulates the feeder system based on the probabilistic optimal power flow method based on the perception of voltage overlimit risk,so as to realize the optimal operation of the feeder system within 99% probability confidence interval.Based on the probability distribution information of distribution network node voltage,this thesis proposes a method to calculate the voltage overlimit risk,and establishes an optimization regulation model of feeder system with the lowest voltage overlimit risk as the objective function.Meanwhile,an improved Latin hypercube Monte Carlo particle swarm optimization algorithm is proposed to solve the optimization problem.Compared with the traditional Monte Carlo particle swarm optimization,this method has faster solving speed and convergence speed,and has strong practicability.3)Considering the station area system lack of measuring devices makes power information can not be collected in real time,which leads optimal power flow method can not be used to optimize the area voltage,based on the deep reinforcement learning method to build the station area optimization regulation structure,realizes the partially visible station area voltage deviation multi-objective optimization with minimum carbon emissions.Based on the historical measurement data,the offline training of deep Q network is completed in the process of continuous interaction with the simulation environment.In the practical application process,only part of the collected node power data of the station area is taken as the network input,and the deep Q network can output the optimization regulation strategy of the current state.