s Transformer Fault Diagnosis Based on Multi-Algorithm Fusion

Background: To make up for the deficiency existing in single method for transformer fault diagnosis, a model of multi-algorithm fusion based on improved Dempster-Shafer (D-S) evidence theory was proposed through analyzing the implementation process of quantum particle swarm optimized BP neural network (QPSO-BP). Methods: According to the failure modes of transformer, the primary fault diagnosis was achieved using a model group formed by several single methods, such as QPSO-BP, the inertia weight PSO optimized BP network (IWPSO-BP) and the constriction factor PSO optimized BP network (CFPSOBP), then the fusion decision was implemented by D-S theory. In view of the defect of standard D-S which can not synthesize the highly conflicting evidences, the credibility factor was used to improve the capability of information fusion. Results: Diagnostic results show that, compared with the single models and standard D-S, the proposed method has stronger fault tolerance, and improves the accuracy of transformer fault diagnosis. Conclusion: The method based on the multi-algorithm fusion can enhance effectively the diagnostic efficacy, and suitable for the pattern recognition of transformer fault.