Prediction Of Line Loss In 10kV Distribution Network Based On Grey Relational Analysis And Improved Neural Network

With the rapid development of the power system,more and more attention is paid to thr reseach of Loss reduction and energy saving.Line loss rate is an important comprehensive technical and economic indicators which can measures the operation and management level of electric power enterprises.It is an essential job to pay efforts to reduce power loss in power grid for power supply enterprises at all levels.the line loss of 10 kV distribution network has accounted for more than 20%in the total power network currently,it has a large loss reduction space.However,Because of the huge structure of the medium voltage distribution network and the large number of components and nodes,it is difficult to collect some necessary operation data,which results in inaccurate calculation results of line losses.Based on the problems above,it is of great significance to put forward a fast and accurate method for predicting line loss of 10kV distribution network.In order to evaluate the line loss level of 10 kV distribution network more accurately and effectively,a method of line loss prediction for 10 kV distribution network based on grey correlation analysis and improved neural network is proposed.Firstly,density-based data cleaning method is adopted to improve the quality of distribution network data,and the outliers in original data is detected and deleted.Secondly,the grey correlation analysis method is used to analyze the correlation between each electrical index and line loss,and the correlation degree of electrical indexes are ranked.Then combined with the actual distribution network data,the electrical characteristic index system which can best reflect the operation status and the structure of the 10 kV distribution network is built.Thirdly,Considering that the structure of traditional BP neural network(BPNN)is difficult to determine and the training process is prone to fall into local minimum and slow convergence speed,the following two methods are adopted to improve it.One is that cross validation method and trial-and-error method are used to analyze the prediction performance of BP neural network under different network structures,so that best hidden layer nodes can be determined.Another is that adaptive genetic algorithm is used to search the weights and thresholds of BP neural network globally to improve the accuracy and convergence speed of the algorithm.329 lines of a practical 10 kV distribution network are taken as an example to establish the line loss prediction model,And the difference of convergence and accuracy between the proposed method(AGA-BPNN)and particle swarm optimization BP neural network(PSO-BPNN),radial basis function neural network(RBFNN)and traditional BP neural network is compared and analyzed.The results show that the evaluation errors of the four methods are 6.71%,12.38%,12.95%and 17.05%respectively,which proves that AGA-BPNN has better convergence,accuracy and validity.Finally,The line loss of 318 lines in the distribution network is predicted from September to December,and the predicted line loss rates are between 0.9%and 5.1%.,which provides a basis for the formulation of loss reduction measures and does advantage to the safe and economic operation of the power grid.