Research On Classification And Localization Of Transformer Partial Discharge Based On Deep Learning Theory

Partial discharge is a common problem in the operation of power transformers.It is difficult to collect partial discharge fault information of transformers and locate them accurately,which has been a difficult problem in power systems for many years.In this paper,the location of partial discharge in power transformer is studied and the partial discharge model of transformer is established.Through deep learning,the precise location of partial discharge in transformer is realized.The main contents of this paper are as follows:(1)In view of the problem that the electromagnetic wave and ultrasonic data of actual transformer partial discharge cannot be obtained,this paper establishes a transformer partial discharge experimental platform,which selects the transformer discharge error as the partial discharge model to measure and collect the electromagnetic wave and ultrasonic data,providing data support for the subsequent research on transformer partial discharge classification and fault location,and tests the feasibility and effectiveness of the platform through experiments.(2)Aiming at the low accuracy of the traditional classification method of transformer partial discharge based on CNN model,this paper proposes a new classification method based on Res Net and Dense Net models.Compared to traditional CNN models,Res Net and Dense Net models have good performance in resisting noise interference,and can still achieve certain classification accuracy when training data samples are reduced.By analyzing the classification accuracy of the model under different network layers and iteration rounds,it was found that the Res Net and Dense Net models achieved the optimal classification accuracy at 50 layers and 50 iterations.The experimental results show that the classification accuracy of Res Net and Dense Net models in transformer partial discharge classification is more than97%,which is more than 10% higher than that of traditional CNN models.(3)This paper proposes a precise transformer partial discharge fault location model based on graph convolutional neural network and graph attention network to address the issue of existing machine learning models for fault location tasks not considering data correlation.The architecture of the model preserves the correlation between data,learns to integrate data information from multiple measurement units,extracts and combines features layer by layer,effectively improves the accuracy of transformer partial discharge precise location,verifies the accuracy and effectiveness of the proposed method through experiments,and tests the performance of the model under the large node system.The robustness of the model is tested through data disturbance and data loss.Experiments show that the proposed model meets the task requirements of accurate location of transformer partial discharge.