A Power Transformer Event Classification Technique Based on Support Vector Machine

Currently, different artificial intelligence techniques have been applied in power transformer protection purposes to discriminate internal faults from inrush currents and other disturbances. This paper proposes the application of a support vector machine (SVM) algorithm, for distinguishing among internal faults, external faults, and transformer energizations in a power transformer. The event classifier is enabled through a disturbance detector, hence receiving as input features the first post-fault boundary wavelet differential energies, which are processed by the classifier during the training and set stages. Simulations of a 100 MVA rated power transformer using the Alternative Transients Program (ATP) were carried out. Hence considering a wide variety of the fault parameters, a performance analysis of the SVM classifier regarding the overall accuracy and the time of operation in discriminating the events was accomplished, and promising results were achieved.