Identifying line-to-ground faulted phase in low and medium voltage AC microgrid using principal component analysis and supervised machine-learning

Abstract

A supervised machine-learning based approach for faulted phase identification in bolted, low- and high-impedance line-to-ground faults using principal component analysis for feature extraction from multiple input signals is presented in this paper. DIgSILENT PowerFactory is used for simulating the underlying microgrid to obtain fault related data, while MATLAB is used for machine learning application. A 15-fold cross validation is applied to the training dataset for evaluation of different machine learning models and the results show supreme performance compared to previous methods.