Research on Relevance Vector Machine and Its Application to Fault Diagnosis of Transformers

Transformer fault diagnosis based on relevance vector machine (RVM) is proposed. The advantages of the RVM over the support vector machine (SVM) are probabilistic predictions, automatic estimations of parameters, and the possibility of choosing arbitrary kernel functions. Most importantly, RVM is capable of comparable classification accuracy to SVM, but with fewer relevance vectors (RVs) and higher testing speed. Performances of RVM are analyzed and validated using typical classification examples and then RVM is applied to fault diagnosis of transformer. The RVM-based fault diagnosis of transformer is described in detail. The method takes normalized transformer feather gases content as inputs. Transformer fault diagnosis model is constructed based on binary tree classification method. Experimental results show that RVM-based transformer fault diagnosis is capable of comparable or even more excellent diagnosis accuracy to SVM, but with typically highly sparse models and highly diagnosis speed.