Analysis of Machine Learning Methods for Fault Classification in Electric Systems with Distributed Energy Resources

The continued growth of Distributed Energy Resources (DER) presents challenges to electrical grids, not the least of which is electrical protection. This article develops a classification criterion to identify the fault zone in a distribution network using some well-known machine learning methods on transient RMS signals. The signals considered are: active power, reactive power, voltage, current and frequency extracted from the Power Plant Controller (PPC) of the distribution generator inverter. Then, this information is part of the input for different classification methods to evaluate their efficiency, this by simulating different fault cases. The simulations were performed on a generic model obtained from Power Factory, while the analysis for testing and training was performed in Matlab. The results show that the multilayer perceptron method gives better results.