Research On Transformer Fault Diagnosis And Prediction Based On CVA

Oil-immersed transformers are one of the hub devices in the construction of smart grids and interconnected power grids.Fault diagnosis of transformers based on oil chromatographic analysis technology(DGA)is the development direction of the safe operation of power grids.Transformer fault prediction is an important part of transformer maintenance.Accurate fault prediction can eliminate transformer faults in advance by taking appropriate measures before transformer faults occur,and reduce the failure of the power system due to its faults.Economic loss.This paper conducts a systematic study on transformer fault diagnosis methods and prediction models,and proposes a fault diagnosis method based on wavelet neural network combined with canonical variable analysis(WPTCVA),particle swarm optimization support vector machine(PSO-SVM)and wavelet transform time The sequence(WPT-ARMA)combination forecasting model improves the accuracy of fault diagnosis and the prediction accuracy under a small sample of data.The research content of oil-immersed transformer in this paper is divided into the following parts:First,analyze the common fault types of transformers,and divide transformer faults into electrical faults and thermal faults according to the nature of the fault.Through the summary of intelligent algorithms and traditional algorithms,the fault diagnosis algorithm is determined to be a wavelet neural network algorithm,with a combination prediction model of support vector machine model(SVM)and autoregressive moving average model(ARMA).Secondly,a brief introduction to the wavelet neural network,the use of wavelet neural network algorithm to establish a transformer fault diagnosis model,determine the input layer,hidden layer and output layer,and then select a large number of representative data samples,use wavelet transform for noise reduction.On this basis,the sample data is normalized,and training,testing and error analysis are carried out.It is concluded that the convergence speed of wavelet neural network is fast but the convergence error is large.Third,the wavelet analysis combined with CVA method is used to process the gas data dissolved in the transformer oil,not only does not need to manually add optimization coefficients or penalty factors,but also the transformer fault diagnosis is fast and accurate,which is better than the current diagnosis accuracy rate.The smart way.By separating the collected 400 DGA data sets into 275 training sets and 125 test sets,the fault diagnosis performance of this method is evaluated.After training,the test results show that the diagnostic accuracy rates of the WPT method and the CVA-based method are 90.4 and 93.4%,respectively.From these data,it can be clearly seen that the diagnostic rate of this method in classifying transformer faults has increased by 3%.Finally,through the normalization of DGA gas data,kernel function selection,and particle swarm optimization support vector machine to establish a PSO-SVM prediction model,wavelet denoising,model identification,and parameter estimation of DGA gas data to establish a WPT-ARMA prediction model Based on the two prediction models,a combined model was established to achieve the smallest error.Finally,through example verification and comparison,it was found that the prediction model error of WPT-ARMA was smaller than that of the PSO-SVM prediction model,and the combined prediction model could integrate multiple single The information contained in the forecast model effectively reduces the forecast risk.