Research On Technology Of Transformer Fault Diagnosis Based On Improved Neural Network And Ratio Method

Using artificial intelligence methods such as neural network to deeply learn and analyze big data of power equipment is the development trend of intelligent operation and maintenance of power equipment.At present,most of the oil-immersed transformers operated in the power grid can analyze their running state by detecting the gas content in oil through dissolved gas analysis(DGA).On the basis of analyzing the traditional and intelligent transformer diagnosis methods based on DGA,this paper studies a new transformer fault diagnosis technology based on the fusion of improved neural network and ratio method,which considering their respective strengths and weaknesses.This technology screens out the samples that may be misclassified by neural network in advance and transfers them to the traditional ratio method for single data diagnosis,so as to achieve the purpose of single sample error correction,improve the diagnostic accuracy and make neural network play a better role in transformer fault diagnosis.Aiming at the blindness of initial assignment in the process of sample generation by traditional variational auto-encoder,a new strategy of sample generation based on improved variational auto-encoder is proposed for transformer fault data enhancement.In order to make the deep neural network adapt to the characteristic of few types of characteristic parameters of DGA data,an improved one-dimensional convolution neural network is built as the basic classifier of fusion classification technology,and the classification results are output in the form of probability.On the basis of the classifier proposed in this paper,a new technology of transformer fault diagnosis based on the fusion of improved neural network and ratio method is studied.The fusion mechanism and specific flow of the proposed method are given,and how to apply it in transformer fault diagnosis is introduced in detail.The performance of the data enhancement method and the basic classifier proposed in this paper are tested and compared to verify the feasibility of the two methods.The performance of transformer fault diagnosis technology based on fusion classification method is tested and compared.Its adaptability and optimization effect in different basic classifiers are tested respectively,and the influence of different data sets on fusion classification is compared.The results show that both the data enhancementmethod and the basic classifier proposed in this paper have better performance than similar models.The fusion classification method can improve the performance of the basic classifier in different data sets,and the improvement effect is stable for the basic classifier with higher initial accuracy.Figure [20] Table [15] Reference [70]...