| Abstract: |
Modern power grids are now equipped with Pharos Measurement Units (PMUs) sensors that provide real-time, high-resolution measurements of transmission line behaviour like never before. This paper discusses an extensive empirical analysis of artificial intelligence powered Fault detection and classification frameworks using real time PMU data streams from 400 kV and 220 kV transmission corridors in the Indian power grid. A hybrid Convolutional Neural Network (CNN), Long Short-Term Memory (LSTM) networks, and a Random Forest classifier-based fault detection system is proposed in which overall detection of accuracy of 98.76% and classification of the accuracy of 97.43% are achieved against ten random selected fault categories SLG, LL, DLG and all three-phase (3P) faults classifiers. We used a dataset of 52,800 fault event records from fourteen substations which have been monitored via PMU in a given area over a window of twenty-four months. Model robustness and generalizability over the conclusions were confirmed with statistical validation using one-way ANOVA (F = 142.67, p |
