| Urban power grid is an important part of power system.It is the "last kilometer" connected to users in the power grid,which directly affects the power supply reliability and quality of users.In case of out of limit,maintenance and fault of power grid,it is necessary to transfer part of the load to other nodes through load transfer operation.However,due to the complex structure and large number of nodes in the distribution network,it is difficult for dispatchers to give the optimal real-time load transfer scheme.At the same time,the uncertainty of topology change caused by the lack of measuring devices,the random fluctuation of load and the uncertainty of new energy output have brought new challenges to load transfer.Therefore,it is urgent to study the generation algorithm of urban power grid load transfer scheme in random environment to improve power supply reliability and user satisfaction.Aiming at the randomness of power grid topology,this paper carries out urban power grid topology identification.Based on the identified real-time topology,the load transfer path is generated from the perspective of model driven and data driven.In the aspect of model driving,the main network and distribution network are searched together to generate the transfer path,and the path is comprehensively evaluated under the premise of uncertain power grid parameters.In the aspect of data-driven,the load transfer behavior modeling is completed,and the load transfer algorithm architecture based on reinforcement learning is proposed.After the agent obtains the real-time state of the power grid,it can immediately give the optimal transfer path.Firstly,the grid topology identification in random environment is carried out.The joint probability model of topology identification is established based on Markov random field after node voltage raw data processing,and the regularization idea is introduced to avoid the problem of over fitting.The long-term and short-term memory particle swarm optimization algorithm is used to solve the model,and the real-time topology of power grid is produced based on the minimum spanning tree,which lays the foundation for the following formulation of load transfer scheme.Secondly,from the perspective of model driven,the collaborative transfer path search of main network and distribution network is carried out.Based on the real-time topology of power grid,a topology modeling method between generators,network and load is proposed.The set is used to replace the adjacency matrix representing the topology,and the modeling method and data storage structure of different voltage level topologies are given.Based on the topology model,the collaborative path search method of main network and distribution network is proposed,and the path search algorithms in stations and between stations are given.The idea of edge matching is proposed to carry out cross system line matching and realize cross system path search.Then,the searched load transfer path is comprehensively evaluated in a random environment.Based on the stochastic modeling of load and new energy in urban power grid,Monte Carlo simulation is used for probabilistic power flow calculation.The load transfer evaluation index reflecting the change of path itself and power grid state is proposed,and the entropy weight method is used to assign the index weight to realize the evaluation of transfer scheme in probability scenarios.Finally,from the perspective of data-driven,the generation of load transfer path is realized based on reinforcement learning.The load transfer process is modeled based on Markov decision process,and the action space,objective function and constraints of load transfer are defined.Based on DDQN algorithm,the generation architecture of transfer path is constructed,and the trigger conditions of the algorithm are proposed to avoid frequent topology adjustment.Using the thought of exploration and guidance,the movement selection strategy during training is improved.After the trained agent obtains the real-time state of the power grid,it can generate the optimal transfer path immediately. |
