Power Transformer Fault Diagnosis Based On Extreme Learning Machine

Power transformer fault diagnosis using DGA has the feature of detectingpotential transformer failure and can be carried out during runtime. It is a ofSignificance of study and an improvement of operation and maintenance level of thepower transformers. Based on the analysis of the characteristics and shortcomingsof the existing diagnosis methods, The extreme learning machine (ELM) was firstlyapplied to the fault diagnosis of oil-immersed power transformer in this paper.An ELM-based was proposed in this paper. The affection of hidden layeractivation function on diagnostic performance was investigated, and theimplementation procedure of fault diagnosis was provided in detail. The proposedELM-based diagnosis method has the feature of not prone to have the local value,fast training, simple configuration, especially suitable for online diagnosis. Itsdiagnosis performances were validated by case studies.A fault diagnosis method for oil-immersed power transformer based onmulticlass WELM was proposed. This method was proposed to solve the problem ofdata imbalance in DGA by using weighting scheme. Study of different weightingscheme was drawn. Performances of power transformer fault diagnosis based onWELM were validated by case studies.Method of power transformer fault diagnosis using KELM was proposed basedon the study of KELM parameter optimization. A KELM’s parameter optimizationmethod combining the PSO and K-fold validation is proposed. Experiments showthat, comparing to SVM, the KELM transformer fault diagnosis has more accuracyand less training time.