Application of Machine learning algorithms for Power transformer Internal faults identification

摘要

Power transformer insulation failure is one of the key concerns for the effective operating state of transformers. The breakdown of insulation due to exposure of systemic faults for a long- duration lead to inter-turn short circuit of windings, discs, and movement of winding in the radial and axial directions are considered here for evaluation. The analysis with traditional methods needs expertise which may not generate comprehensive results. This paper describes an approach to extract significant features of impulse test responses of the transformer using continuous wavelet transform in the time-frequency domain and extracted data is balanced with random sampling. The performance of three types of classification algorithms is compared to classify (or) predict the fault current data. The comprehensive results of training accuracy indicate that the ensemble algorithm has been performed best among the other classifiers and the prediction rate of the ensemble classifier is represented by the probability curve and the prediction accuracy of the classifier algorithms was presented based on the true positive rate