Evolutionary Bayesian Fusion for Transformers Fault Detection

This paper presents an evolutionary Bayesian fusion method for transformer fault detection. It adopts Bayesian Network (BN) to explore the causal relationship between different potential faults inside transformer and fault symptoms; and then such knowledge is used to identify the fault types of transformer. Since Bayesian network acquires fault evidence gradually and transformer fault diagnosis is also an evolutionary process, the proposed method determines an optimal set of measurements, which need to be performed at each diagnostic step. This methodology can improve the accuracy of transformer fault identification while improve the efficiency of diagnostic process since the number of required measurements is minimized and only meaningful fault evidences are used in fault identification. Case studies are presented to verify the proposed method.