An EKF‐SVM machine learning‐based approach for fault detection and classification in three‐phase power transformers

Abstract

In this paper, a hybrid approach for effective diagnosis of power transformers is proposed. In the proposed method, the extended Kalman filter is used for the estimation of three-phase currents in the primary windings of the transformer. Three residual signals are defined as the differences between the measured and estimated three-phase currents. When the transformer is healthy, the EKF perfectly estimates the primary currents and hence, the residual signals are nearly zero. However, when the transformer is faulty, the EKF cannot suitably estimate the currents due to the large model mismatch resulting from the transformer internal faults. Consequently, large residual signals are generated, which are used as the key signatures for discriminating the internal faults from energisation conditions. Besides, the proposed method uses the entries of the covariance matrix of the estimation error to locate and classify the internal faults. For these purposes, support vector machine classifiers are used. The effectiveness of the proposed method is demonstrated by a large number of simulation test cases obtained using PSCAD/EMTDC software. Also, hardware-in-the-loop experiments are conducted using dSPACE1104 development platform and real-time feasibility of the proposed method is authenticated.