Research On Power Fault Diagnosis Algorithm Based On Optimized Support Vector Machine

The power transformer is an important equipment in the power distribution network of China.Failure early warning and detection efficiency are the greatest guarantee for stable operation of the power grid.Dissolved Gas Analysis(DGA)method was widely used in the production practice.However,in the face of large-scale and high-quality power requirements,the fault monitoring accuracy of this method can no longer meet the needs.This topic was focused on a local oil-immersed transformer that has experienced accidents in recent years.Based on the DGA data,an optimized SVM method was used to divide fault categories,improve the monitoring accuracy of oil-immersed transformers,and ensure normal power supply in specific areas.The research content of the thesis is as follows:(1)The relationship between the oil-gas dissolution principle of the local oil-immersed transformer and the failure of previous experience was studied.The inter-turn discharge fault of the transformer in a local substation was the research object.The problem of poor real-time performance and many misjudgments in the traditional characteristic gas discrimination method were analyzed.According to the divided local fault categories,the method of optimized SVM was used to analyze and diagnose the real-time monitoring data.(2)The min-max normalization processing and PCA feature reconstruction were used as preprocessing methods for DGA data sets.The preprocessing highlighted the min-max function,which increased the identity of the data and improved the diagnostic accuracy.The feature reconstruction selected two methods of multi-dimensional scale(MDS)and principal component analysis(PCA)for comparative to verify the accuracy of fault reconstruction,and improving the accuracy of fault diagnosis.The correlation coefficient of PCA using KMO value was 0.9 better than MDS.Therefore,PCA was selected as the reconstruction of the data.(3)The optimized GWO-SVM and GA-SVM was used to diagnose the DGA data.The appropriate function,selection operator,crossover operator and mutation operator of GA-SVM were selected.The GWO was optimized for defecting to fall into the local optimal loop,to propose an optimized GWO-SVM.Four verification functions were used to verify the computing accuracy of algorithms.The results show that the optimized GWO-SVM was better than the GWO-SVM and GA-SVM.It achieved convergence first in the shortest number of iterations.The diagnostic accuracy of the model was improved by 3% after feature reconstruction.The experimental results of the failure case show that the diagnosis accuracy and training time of the optimized GWO-SVM have been improved.The diagnosis accuracy was 3% higher on average.Improved the ability to diagnose and analyze unexpected faults.