Comprehensive Approaches for the Differential Protection of Power Transformers Using Advanced Classification Techniques

The transformer is one of the costlier and critical assets in the power system. Its protection against internal fault is achieved by implementing a differential protection scheme. However, this scheme holds the false tripping risk subjected to high inrush current drawn during transformer energisation. In the proposed work, wavelet transform and artificial intelligence (AI)/machine learning (ML) based schemes are utilised to differentiate internal fault, normal and inrush current signals. The current signals are generated from the MATLAB simulation software. Later, discrete wavelet transform (DWT) is used to extract essential features from the corresponding current signals. In DWT, the mother wavelet and decomposition levels are selected as Daubechies (Db) and level 4 respectively. The extracted features are then fed to various AI/ML models such as k-nearest neighbors, support vector machine, and a multi-layered feed-forward neural network (MLFFNN) trained by the back propagation and resilient back propagation algorithms respectively. The results of these models are then compared to examine the accuracy of the model at utility/industry-defined sampling frequency.