Fault Detection in Smart Grids Using Deep Learning-Based Phasor Measurement Unit Data Analysis

Abstract

With the rapid growth of smart grids, their operation has become more involved, so new intelligent fault detection systems are needed to maintain grid stability, dependability and strength. Usually, traditional systems have trouble detecting and identifying faults quickly and exactly when the grid is operating under changing conditions and has a high amount of renewables. Because they provide highly detailed and synced measurements, Phasor Measurement Units (PMUs) are now a key tool for real-time monitoring of the grid. In our study, we offer a reliable fault detection method using deep learning which makes use of data from several PMU channels for accurate detection and localization of faults in the system. In the architecture, CNNs are used to find local information from phasor streams and this is followed by sets of BiLSTM layers that model both forward and backward relations in the data related to grid events. A hybrid CNN-BiLSTM model is constructed using data collected from the IEEE 39-bus system that covers many kinds of faults and different levels of noise and loading. The outcomes from experiments show that the presented model does better than traditional Support Vector Machines, Random Forests and k-Nearest Neighbors at classifying faults and at reaction time. With noise and missing information present, the model is able to give highly accurate and comprehensive results. In addition, the framework is fast-responsing, letting it suit real-time use in monitoring over networks. The research results help build intelligent protection systems that can handle issues automatically and quickly, support self-healing of the grid and support future research in maintenance, grid cybersecurity and reserving grid operations with smart analytics.