A Machine Learning Approach for Classifying Faults in Microgrids using Wavelet Decomposition

In this paper, we propose a machine learning approach for fault identification and protection of microgrid circuits. We model a representative microgrid architecture found in an industrial facility in simulation to motivate and demonstrate our approach. In particular, we formulated a 5-class classification problem to identify 5 different fault scenarios. Three-phase voltage and current waveforms corresponding to each fault type were analyzed using 5-level wavelet decomposition. Statistical features were calculated for the resulting wavelet coefficients. Significant features were selected using the Wilcoxon rank-sum test while the classification performances of linear discriminant analysis (LDA), support vector machines (SVM) and Naïve Bayesian classifiers were compared. Experimental results showed that SVM has the highest performance and the 5 fault types can be identified with overall accuracy of 98.76% within 0.15 seconds from fault onset.