Research On Fault Diagnosis Of Power Transformer Based On Deep Learning

The operating status of a power transformer is related to the safety and stability of the entire power system.Therefore,in engineering applications,real-time monitoring of its status is required to detect latent faults in time.At present,the primary method for realizing transformer condition monitoring is Dissolved Gas Analysis(DGA).Aiming at the limitations of transformer intelligent fault diagnosis methods based on DGA,this paper proposed a transformer fault diagnosis method based on improved Sparrow Search Algorithm(ISSA)to optimized Deep Brief Network(DBN)and support vector machine(SVM).The network structure parameters of DBN were optimized with ISSA,so that the relationship between the characteristic information of the DGA fault data and the fault type can be deeply mined.Combined with the advantages of SVM that can effectively solve the classification problem of small samples,a transformer fault diagnosis model combining DBN and SVM was established.The specific research content and results were as follows:(1)A transformer fault diagnosis model based on DGA feature ratio and deep belief network was constructed.The ratio of characteristic gas of transformer fault was deleted by neighborhood rough set.The redundant information between the feature inputs was removed so as not to affect the fault feature extraction effect of DBN.Compared with the current intelligent methods for transformer fault diagnosis,the results showed that the DBN diagnosis model has higher accuracy in predicting the fault state of transformers.(2)An improved sparrow search algorithm was used to optimize the performance of the DBN model.In order to improve the global optimization capability of SSA,dynamic reverse learning strategy and Gaussian mutation were introduced to improve SSA.The algorithm performance test of ISSA showed that the improved algorithm can search the global optimal parameters more quickly and accurately,and has better generalization ability than SSA.Accordingly,a fault diagnosis model based on ISSA and DBN was constructed.ISSA was used to optimize the inter-layer neuron connection weights and neuron bias of the RBM in DBN,which further improves the ability of DBN to extract the essential characteristics of the input data.The ISSA-DBN model was used to predict and classify the fault status.The results showed that the fault diagnosis accuracy rate of this model can reach 94.44%.(3)A transformer fault diagnosis method combining DBN and SVM was proposed.In view of the superior feature extraction ability of DBN and the excellent classification ability of SVM for small samples,an ISSA-based DBN-SVM transformer fault diagnosis model was established.The trained samples were inputed into the DBN optimized by ISSA to fully extract the characteristics of the fault nature.Then the extracted fault features were inputed into ISSA-SVM for classification trained.Through the diagnosis results of the actual operating state of the transformer,the fault diagnosis accuracy of this model was higher than ISSA-DBN and ISSA-SVM,and the prediction accuracy of the transformer operating state can reach 95.56%.(4)The specific cases of transformer faults were used to verify the generalization ability and practicability of the model.The results showed that the DBN-SVM diagnostic model optimized by ISSA extracts the essential characteristics of the fault from the transformer DGA data.So it can accurately diagnose the latent faults in the operation of the transformer.The transformer fault diagnosis model of DBN-SVM optimized by ISSA can used the oil chromatogram data detected on-line to accurately diagnose the latent faults in the transformer operation in time and formulate reasonable maintenance strategies based on this.Furthermore,the operating status of the transformer is monitored in real time to ensure the safety and stability of the power system.