Research On Prediction Of Gas Content In Transformer Oil And Fault Diagnosis Of Transformer

Power transformer is an important component in power system,and its running state plays an important role in the safe operation of power system.Fault prediction and diagnosis of transformer is the basis to ensure its normal operation and implement state maintenance.The content of dissolved gas in transformer oil is an important characteristic index of transformer operation state.According to the historical data of dissolved gas content in transformer oil,it is very important to predict the change trend of the content in a certain period in the future.In this paper,the variation prediction of dissolved gas content in transformer oil and the classification diagnosis problem which may cause transformer fault are studied.This paper presents a prediction method to determine future data based on historical data of gas content.Aiming at the problem of short prediction interval and relatively low prediction accuracy,this method is based on the differential autoregressive moving average model(ARIMA),which is improved in the process of optimal parameter and model testing and screened by three criteria,the accuracy of the model is ensured by four test methods.The series of operations are estimated and modified to obtain the optimal prediction model.According to the relevant guidelines,the data series of five kinds of gas content dissolved in transformer oil are determined.The experimental results show that the prediction effect is good,which can provide valuable reference for reasonable arrangement of transformer condition maintenance.Taking the dissolved characteristic gas content in oil as the characteristic quantity,a fault diagnosis method of transformer is given.This method mainly uses the convolution neural network algorithm principle to construct the diagnosis model,aiming at the problem that traditional neural network is prone to over-fitting,and improves the pooling operation of the model.The maximum dropout pooling method is used to replace the original mean pooling or maximum pooling in the pooling layer,which enhances the generalization performance of the model.Taking the five gas content values dissolved in transformer oil as input vector and the coding values corresponding to the six operating states of transformer as output vector,in the process of model building,the position and number of parameters are analyzed,and the activation function is determined by the method of optimal selection function.The influence of the size and number of convolution kernels,the number of iterations and the number of samples in each group on the accuracy of the model are discussed in detail.Applying the network generated by this method to transformer fault diagnosis not only overcomes the misjudgment caused by insufficient coding and absolute boundary in the traditional three-ratio method,but also improves the accuracy of diagnosis and avoids the shortcoming of overfitting BP neural network.A diagnostic example proves the effectiveness of this method.