Study On Deep Learning Based Approaches For Partial Discharge State Recognition On Transformers

Power equipment is an important equipment related to the national economy and the people's livelihood.The stability of its operation state has a serious impact on the economy and security.So,it is very important for monitoring and fault diagnosis of power equipment.Due to the complex field environment of electric power equipment,the fault diagnose is significantly complex.Current methods include signal processing processes.The feature extraction is accomplished under the direction of experts,which requires abundant electrical knowledge.These methods have poor adaptability.The existence of a large number of random factors leads to failure recognitions.In order to increase the recognition rate,this paper put forward a Faster R-CNN method.This method extract the PD features directly,and then recognize PD pattern automatically.Tests show this method has better performance than traditional methods.Main work:(1)A novel feature extraction method for partial discharge based on cross-wavelet transform was proposed,aimed at the high sensitivity of the traditional feature extraction method to noise.Case analysis demonstrated that the proposed method could effectively avoid the influence of noise.(2)Based on the Faster R-CNN frame,to grasp the process and significance of its structure and the operation of convolution pool and realize the convolution operation of onedimensional time series.The convolutional pooling structure is sparse and is extracted to more detailed characteristics.(3)In view of the traditional training methods need a large sample database,corresponding improvements were made on its structure and the way of training,put forward a discriminant sexual characteristics of convolution method of study,more rapid for partial discharge fault characteristics of the study.The recognition of the partial discharge state of Faster R-CNN is used for the diagnosis of partial discharge fault in multiple working conditions,which improves the diagnosis speed and accuracy.