Research On Transformer Fault Diagnosis Based On SMOTE And Grey Wolf Optimization Probabilistic Neural Network

With the rapid development of society and economy,various industries have entered a period of rapid development,which also puts forward higher requirements for the operational reliability of the power system.Transformer as the indispensable core equipment in the power system,to ensure the stable and reliable operation of the transformer is the key to ensure the safe operation of the power system.In the actual application of transformers,if the equipment fails,it will affect people’s production life and even lead to catastrophic events.Therefore,the research of accurate and efficient fault diagnosis for transformers is of practical significance for the reliable operation of power systems.In this paper,starting from the research related to fault diagnosis of neural network,we deeply analyze the fault mechanism as well as types of oil-immersed transformers,and propose a transformer fault diagnosis method based on SMOTE and gray wolf optimized probabilistic neural network,which is mainly as follows:(1)Based on the Dissolved Gas Analysis(DGA)method in oil,a brief summary of the current research status of transformer fault diagnosis is presented.The basic principle of gas composition generated by internal faults in oil-immersed transformers is introduced,and the correspondence between gas content values and fault types at transformer faults is derived through the analysis of different fault mechanisms,which provides a theoretical basis for further research on transformer fault diagnosis.(2)A transformer fault diagnosis method based on gray wolf optimized probabilistic neural network is proposed for the problem that the traditional three-ratio method has fuzzy coding boundaries and incomplete coding leading to poor diagnosis of fault data.Firstly,by studying the principle of probabilistic neural network,the network structure design of the diagnosis model is completed.Then an 8-dimensional data vector based on the three-ratio method is constructed as the fault feature vector of the model,and the model is trained and tested.Secondly,the theoretical analysis of the gray wolf optimization algorithm is carried out,and the algorithm is improved to improve the performance of the algorithm for the problem that the gray wolf optimization algorithm tends to fall into local optimum.Finally,the improved gray wolf optimization algorithm is used to optimize the parameters of the smoothing factor of the probabilistic neural network to reduce the influence of human factors on the diagnostic performance of the model.The experimental results show that the strategy of optimizing the proposed probabilistic neural network model using the improved gray wolf algorithm is beneficial to the improvement of the model diagnosis accuracy.(3)Aiming at the problem of less fault data and unbalanced number of fault categories in the actual operation of the transformer,the transformer fault diagnosis model has unbalanced diagnostic accuracy of each category.A transformer fault diagnosis method based on SMOTE and gray wolf optimization probabilistic neural network is proposed.Firstly,SMOTE algorithm is used to sample the unbalanced transformer fault data,and then through visual analysis,the difference of sample data distribution before and after sampling is evaluated.Finally,comparative analysis is carried out at the data level and the algorithm level.The results show that,at the data level,the model diagnostic performance of the data set after being equalized by the SMOTE algorithm,compared with the original data set and the data set processed by random oversampling,the diagnostic performance is the best;at the algorithm level,the balanced data set is in gray The comparison experiments of four models of wolf-optimized probabilistic neural network,improved gray wolf-optimized probabilistic neural network,sparrow-optimized probabilistic neural network and seagull-optimized probabilistic neural network,the results show that the proposed improved gray-wolfoptimized probabilistic neural network model can be integrated in transformer fault diagnosis.Best performance.Combining the two aspects,it is proved that the proposed model shows excellent performance on the task of transformer fault diagnosis,and can effectively identify transformer fault types.