Deep CNN with MLP for High Impedance Fault Classification in Power Grids with Distributed Generation

Power demand is increasing day by day for continuing the industries, commercial loads, and household loads due to the increasing population, and it requires consistent power to sustain life. Integrating renewable energy sources into the grid is increasing tremendously to meet the power demand. The fault is one barrier in the power system that interrupts the smooth growing country. High impedance fault (HIF) classification is the main challenge in power system distribution networks due to the low magnitude of fault current. In addition to having low current, HIFs have other typical properties comparable to the characteristics of switching events, like load switching, such as unpredictability, nonlinearity, etc. Adding distributed generators (DGs) to the distribution system modifies the substation’s current contribution and impacts HIF detection. In this paper, fault classification has been done, including HIF using a Deep Convolutional Neural Network (CNN) with Multi-Layer Perceptron (MLP). A distribution network integrated with different distributed generators like solar PV panels, Doubly fed induction generators (DFIG), and Diesel generators. Various possible configurations with different inception angles, fault distances, types of faults, and HIF are created in data generation. The weighted average accuracy for fault classification is 99.46% achieved through 10-fold cross-validation of the proposed Deep CNN with MLP model. Various comparisons have been made in the discussion section of the proposed model with existing deep learning models.