Transformer Fault Diagnosis Based On Support Vector Machine And Chemical Reaction Optimization Algorithm

Power transformer is the core equipment in power system,which can change the voltage to achieve efficient transmission of electricity.Its running state is closely related to the security and stability performance of the power system.Once there is a serious fault in power transformer,it can not only bring huge economic losses to the Power Grid Corp,but can also cause inconvenience to the users,resulting in adverse social impact.Finding the latent fault of transformer in time and making the maintenance plan can effectively prevent the deterioration of the fault and reduce the losses.Therefore,the research of transformer fault diagnosis technology is of great significance to improve the reliability and management level of power system.This paper introduces several methods of transformer fault diagnosis and analyzes the diagnostic performance according to different detection signal.Dissolved Gases Analysis(DGA)has been placed in an important role in the preventive tests of power transformer.The DGA technology has been considered as reliable method to diagnose transformer because it includes rich fault feature information.This paper uses the DGA data as the original sample,then transformer fault diagnosis is similar to a nonlinear,high dimension,small samples pattern recognition problem,so we introduced Support Vector Machine(SVM)to construct multi-classification model to identify fault samples.The feasibility and effectiveness of SVM applied in transformer fault diagnosis are verified by using the Libsvm toolbox for case analysis.This method avoids the problems of missing code and absolute encoding ways in traditional diagnosis method like the improved three-ratio method.In order to further enhance the Support Vector Machine classifier for fault diagnosis accuracy and efficiency.This paper uses Chemical Reaction Optimization algorithm(CRO)to search the penalty factors and kernel parameters of Support Vector Machine.Then 3 standard test data sets are used for example analysis,the test results show the performance of CRO algorithm in SVM parameters optimization is more efficient and reliable than Particle Swarm Optimization algorithm and Genetic algorithm.Further more,the Chemical Reaction Optimization Support Vector Machine(CRO-SVM)method is applied to diagnose transformer faults,the simulation results indicate that its diagnostic accuracy compared with the standard SVM method is improved obviously,while its running time compared with the other two SVM optimization methods is decreased dramatically.This research provides a new idea for the intelligent diagnosis of power transformer,which gives guidance for the subsequent maintenance decision.