Automatic Detection of Natural Hazard-Induced Power Grid Infrastructure Faults Using Computational Intelligence

Power grid infrastructure systems are vulnerable to natural hazards. Many studies have focused on the use of sensing technologies to detect natural hazard-induced power grid faults. However, a massive sensor network to collect data for such studies is costly and may not capture complex grid conditions. Therefore, this paper develops an automated grid fault detection system. First, a smart solar-enabled microgrid was developed to simulate small grid operation, which can also dynamically sense the voltage and current for capturing the grid conditions. Second, three types of faults (i.e., partial shading fault, three phase fault, and tripping fault) were introduced into the microgrid to represent the potential faults caused by natural hazards. Third, a one-day operation was simulated. Fourth, a dataset with 864,000 samples was collected, denoised, labeled, and used to develop three different machine learning classifiers. These classifiers were evaluated using four metrics, including accuracy (i.e., the proportion of correct predictions made by a classifier), precision, recall, and F-1 score. Model evaluation results showed that (1) the K-nearest neighbor was the optimal classifier to detect a partial shading fault with an accuracy of 99.19%, and (2) decision tree was the most performant model for detecting three phase fault and tripping fault with accuracies of 100% and 99.90%, respectively. Ultimately, this paper contributes to the body of knowledge by integrating power grid simulation and machine learning for improving the resilience of power grids against natural hazards.