A Real-Time Fault Localization in Power Distribution Grid for Wildfire Detection Through Deep Convolutional Neural Networks

Abstract

Fault localization in power distribution systems plays an important role in the resiliency of large-scale power distribution grids since its accuracy can influence the restoration time that needs to be minimized. In this article, we proposed a real-time deep learning algorithm to classify and localize the faults that occurred in the system based on measured data. To get the accurate fault location that may cause the potential fire, we extracted the feature vectors with the measured values of the distribution lines and applied a deep convolutional neural network classifier on them. Also, the presented method was able to detect the fault type, and compute its released energy that can contribute to wildfire based on the measured voltages and currents of the distribution nodes. Unlike prior data-driven methods, the proposed classifier was based on the single observability of the sensor measurement devices due to the limited access to the data and a limited number of measurement units. The performances of our scheme were validated through simulations of four types of faults in a real distribution test feeder under varying system observability, measurement attribute with and without distributed generation units. The proposed classifiers were able to determine the possible fire due to different fault types under 100% accuracy and find the possible fire location with an average accuracy of 85%.