Hybrid RVM–ANFIS algorithm for transformer fault diagnosis

Dissolved gas analysis (DGA) is a popular method for diagnosing faults inside power transformers. However, some of the recorded data for the analysis are with ambiguous characteristic, leading to misdiagnosis of conventional methods. In this work, a hybrid method, which combines the relevance vector machine (RVM) and the adaptive neural fuzzy inference system (ANFIS) has been proposed to address this issue. Given the fuzziness between DGA records and fault type, and to minimise the number of rules that ANFIS needs to extract, the RVM algorithm performs binary separation firstly, and then ANFIS is utilised to achieve further fault diagnosis in this study. The experimental results demonstrate that the hybrid RVM–ANFIS algorithm can achieve an accuracy rate as high as 95%. Moreover, the proposed algorithm exceeds single ANFIS, support vector machine, and artificial neural network on distinguishing multiple faults and samples with ambiguous characteristic. The engineering application results also demonstrate the effectiveness and superiority of the proposed RVM–ANFIS.