Fault diagnosis method for transformer based on NCA and CapSA-RELM

The key to intelligent fault diagnosis for transformer is to find relevant dissolved gas characteristics with the capability of representing different types of faults. However, the existing problem is that using a single traditional ratio method as the inputs of classification module cannot obtain high classification performance. Therefore, a fault diagnosis method for transformer based on neighborhood component analysis (NCA) and capuchin algorithm (CapSA)-regularized extreme learning machine (RELM) is proposed. Firstly, 21-dimensional features of the oil-immersed transformer are extracted based on the traditional ratio methods such as IEC three-ratio and Roger’s ratio. Secondly, The NCA is used to select the feature to obtain an effective 6-dimensional feature. Then, these features are utilized as the input of the RELM model for training, and the parameters of model are optimized by CapSA. Finally, the feasibility and validity of the proposed method are verified by actual data sets collected from previous publications. The experimental results show that the proposed method has higher diagnostic precision and stability.