Distribution Network Fault Identification Based On Graph Neural Networ

As the demand for electricity in people’s daily production and life is growing,the density and complexity of the distribution network is also increasing.Faults in the distribution network can affect the normal power supply of the network,which in turn affects the normal production and life and causes huge economic losses.How to ensure that failure data is robust and reliable and how to make full use of fault information to quickly and accurately identify conventional faults and leapfrog tripping accidents is of great significance to ensure the safety and reliability of the distribution network.Firstly,in order to solve the problem of missing fault data in distribution networks,the method and process for reconstructing missing fault data based on Graph Autoencoder(GAE)are introduced.The paper describes how to fit distribution network measurement data with GAE and how to solve the optimal completion value of the missing measurement value by constraint search.The results of the experimental tests show that the model can combine the topological information of the distribution network to better learn the intrinsic correlation between each measurement data and improve the accuracy of the reconstructed data compared with the traditional missing data reconstruction method.Subsequently,in order to improve the limitations of Graph Neural Network(GNN)in distribution network fault identification,the method and process of soft label based graph neural network for fault identification.The paper describes how to unify the labeling methods of bus faults and line faults by soft labels and to effectively represent the faults occurring on the additional lines.The experimental test is carried out in this paper,and the results show that the model can expand the range of distribution network fault locations that can be identified by GNN,and also help to improve the adaptability of GNN to cope with changes in distribution network topology.Finally,in order to solve the problem of leapfrog tripping fault identification in distribution networks,the method and process of leapfrog fault identification based on Dynamic Graph Neural Network(DGNN)are introduced.This paper introduces how DGNN can effectively mine the characteristics of leapfrog faults by capturing the continuous action of the protection device and the sequential changes of the topology and electrical quantities.Experimental tests are conducted in the paper,and the results show that the model is able to achieve effective identification of leapfrog faults in a variety of scenarios.