Research On Topology Generation And Reconfiguration Of Distribution Network Based On Machine Learning

With the development of smart grid,devices such as smart meters are connected to the distribution network,which collect the massive operational data in real time.The data contains enormous value to be developed.In addition,the distribution network more frequently interacts with the external equipment such as the distributed power supply and energy storage.The new operation mode seriously interferes with the traditional physical characteristics of the distribution network.The traditional physical theoretical methods are difficult to meet the requirements of the new complex system.Therefore,the new schemes are needed to cope with complex environment of distribution network.New technologies such as machine learning can exploit the potential value of massive operational data and explore new cognitive relationships,which assist existing methods to optimize the operation.Topology analysis,which includes topology generation and reconfiguration optimization,are the basic research contents for the safe and stable operation of the distribution network.The machine learning methods are first explored to solve these core problems in topology analysis.1)In order to accurately monitor the real-time topology of the distribution network,the topology generation algorithm based on Lasso and its supplementary rules is proposed.First the correlation coefficient matrix is calculated by the Lasso model.Then the matrix is modified by the "and" rules and supplementary criterion,and finally the topology is generated through the accurate matrix.The generation error rate of the un-loop or loop topology of 119-bus and other lower complexity network is lower than 6.14%when 720 voltage sampling points are used,and decreases with the increase of voltage points.The algorithm has higher accuracy performance than traditional Chow-Liu and Lasso +"and"algorithms.2)In order to optimize the topology of distribution network efficiently and economically,the dynamic topology reconfiguration algorithm based on LSTM prediction mechanism is proposed.First the LSTM model is constructed to predict the load data of each node in each time period.Then the optimized BPSO model is used to generate the topology reconfiguration plan of each period with the forecasting data,and finally the global optimization dynamic reconfiguration strategy is proposed based on the criterion of the optimal comprehensive cost.The performance of LSTM model is tested with real data of the distribution network in England from 2004 to 2009.The results show that the mean absolute error of prediction is 1.59%,and more than 80%of the rate is distributed in[0,2%],and the distribution is concentrated at different time periods.Irn addition,comparative experiments show that the prediction performance of LSTM model is better than that of traditional artificial neural network and support vector regression model.The performance of the algorithm is tested by 33-bus simulation.The final strategy is to re-divide the operation period into 7 periods and perform 20 switching operations,which reduces the cost of line loss by RMB 1152.42 yuan and improves the voltage quality.The analysis proves that the algorithm has better performance than the traditional online calculation algorithm.The experimental results show that the topology generation algorithm has high accuracy and low computational complexity,which can be employed to monitor the topology structure for assuring the accuracy and security of the operation.Topology reconfiguration algorithm can provide more time margin of reconfiguring and reduce more line loss cost,which can be used to optimize the topology structure for assuring reliability and economics of operation.Without adding new equipments,these algorithms can assist traditional methods to improve the intelligence degree of the distribution network,which has certain exploratory and practicability.