Research On Transformer Fault Diagnosis Based On OFC-DCAE Model

Transformer as an important equipment in power system,the operation status of transformer is related to the stability and security of power grid.Once the transformer breaks down,the economic loss and social impact are incalculable.It is of great significance to improve the safe and stable operation of the power grid to accurately identify the transformer fault types through reasonable and effective methods and provide maintenance assistance to the workers.It is found that most of the fault samples obtained in the case of transformer failure are not labeled,while the traditional diagnosis algorithm can`t make full use of the unlabeled sample data.Aiming at this characteristic,this paper is given priority to with deep learning contractive auto en-coder network,combining with the feature optimization to improve contractive auto en-coder network.This paper proposes based on the combination of genetic algorithm optimization and deep learning contractive auto en-coder network transformer fault diagnosis method.It`s used to improve the utilization rate of the fault samples and the accuracy of fault diagnosis.The main work of this paper is as follows:(1)The optimization of the optimal feature combination.Using 8kinds of features and two ratio form 8 kinds of feature combine 36 group feature,based on the genetic algorithm combine code for nuclear parameters of support vector machine(SVM)、punishment factor of SVM and 36 group feature.According to the cross validation and accuracy of the method accuracy to select a group of optimal feature combination(OFC).(2)Construction of transformer fault diagnosis model based on feature optimization and deep learning contractive auto en-coding network(DCAE).The AE layers of the deep learning contractive auto en-coding network model were determined by using the typical data set.In addition,in order to improve model accuracy and sample data utilization,an OFC-DCAE transformer fault diagnosis model was constructed by taking a group of optimal feature as the input of deep learning contractive auto en-coding network.(3)Test and analysis the OFC-DCAE model.Through the experimental comparison and analysis of factors affecting the performance of OFC-DCAE model,a group of optimal parameters of OFC-DCAE model were determined.Through experimental comparison and analysis,the fault diagnosis accuracy of the model OFC-DCAE,genetic algorithm optimization support vector machine(GA-SVM),particle swarm optimization support vector machine(PSO-SVM),back propagation neural network(BPNN)and deep learning stack auto en-coder network(SAE)were tested.The results show that the OFC-DCAE model is superior in fault diagnosis accuracy and fault sample utilization.In addition,a transformer fault diagnosis system is developed based on OFC-DCAE model.