High impedance fault discrimination in microgrid power system using stacking ensemble approach

Abstract

High impedance (HI) faults in microgrid (MG) power systems are non-linear, intermittent, and have low fault current magnitudes, making them challenging to detect by typical protective systems. Consequently, it is imperative to implement a sophisticated protection system that is dependent on the precision of fault detection. In this study, a stacking ensemble classifier (SEC) is proposed to discriminate HI fault from other transients within a photovoltaic (PV) generated MG power system. The MG model is simulated with the introduction of faults and transients. The features of data set from event signals are generated using the discrete wavelet transform (DWT) technique. The dataset is used to train the individual classifiers (Naïve Bayes (NB), decision tree J48 (DTJ), and K-nearest neighbors (KNN)) at initial and meta learner in the final stage of SEC. The SEC outperforms other classification methods with respect to accuracy of classification, rate of success in detecting HI fault, and performance measures. The outcomes of the classification study conducted under standard test conditions (STC) of solar PV and the noisy environment of event signals clearly demonstrate that the SEC is more dependable and performs better than the individual base classification approaches.