Research On Fault Diagnosis And Prediction Of Power Transformer Based On Support Vector Machine

As the hub electrical equipment of the power system,the power transformer plays an important role in power transformation,transmission and distribution,and its operating status is directly related to the stability and safety of the power grid.It is of great significance to study the technology of power transformer fault diagnosis and prediction,accurately judge the nature of the fault and predict the existing potential faults.According to the characteristics of power transformer faults,the dissolved gas content in the oil is used as the feature quantity,the support vector machine(SVM)theory is used as the basis and other intelligent algorithms are combined,a power transformer fault diagnosis and prediction model based on SVM is established to solve the problem of transformer internal fault diagnosis and operation status trend.The parameters of SVM directly determine the performance of the model.Aiming at the difficulty of selecting SVM parameters,the artificial fish swarm algorithm(AFSA)is introduced to optimize the parameters.Based on the theoretical study of AFSA,the effects of step and visual on AFSA are analyzed.At the initial stage of optimization,the step and visual of artificial fish are expected to have larger values,and in the later period,the step and visual of artificial fish are expected to have smaller values.Aiming at the shortcomings of fixed step and visual in AFSA,an adaptive artificial fish swarm algorithm(AAFSA)is proposed.The experimental results show that the AAFSA has a better optimization effect.The error correcting output code support vector machine(ECOC-SVMs)model is introduced to solve the multi classification problem of power transformer fault diagnosis.The coding matrix is pre-constructed and not combined with specific applications.To solve this problem,a clustering coding matrix is proposed and its effectiveness is verified by experimental comparison.The clustering encoding matrix has a fixed and short code length and does not have redundancy.To address this problem,combining the clustering encoding matrix with the 1-v-1 encoding matrix and appropriately pruning the encoding,a mixed encoding matrix is proposed.The fault diagnosis model of power transformer is established by the mixed coded ECOC-SVMs,which is proved to be effective.For the fault prediction of power transformers,the AAFSA-SVM model of AAFSA optimization SVM is established.The good effect shows that it is feasible to apply the SVM to the dissolved gas prediction in transformer oil.The single-variable grey prediction model(GM(1,1))and the multi-variable grey prediction model(MGM(1,n))are used for prediction and comparison.It is verified that there is a certain correlation between the gases in transformer oil.For the background value problem of the MGM(1,n)model,the AAFSA is used to optimize the background value,and the improved MGM(1,n)model is applied to establish the transformer fault prediction model.The risk of single model prediction is larger,and the combined forecasting model can reduce the risk of prediction.Combining AAFSA-SVM model with improved MGM(1,n)model,a combined prediction model of power transformer fault prediction is established,which is proved to be better than single prediction method.