Fault Localization in Power Distribution Networks Using FLSO-SVR: A Data-Driven Machine Learning Approach

Abstract

Distribution lines play an essential role in the modern power grid and have the potential to influence thereliability of the power supply. Immediate fault localization is essential for a reliable power systemprotection system. When diagnosing a problem, there are two stages. One area that has already reachedremarkable accuracy rates is fault categorization. Fault localization in power distribution networks refersto identifying the exact location of faults or disturbances within the grid. Thus, fault location is the focusof this research, which is the inverse of the original goal. The proposed approach utilizes a combinationof Principal Component Analysis (PCA) for feature reduction, Wavelet Transform for signaldecomposition, and a novel Fuzzy-Logic Spider Optimization-based Support Vector Regression (FLSOSVR) model for accurate fault localization. The algorithm determines the approximate location of the faultby examining voltage and current readings taken at the main feeder the monarch, as well as scheduledinjections of reactive as well as active electricity by networks synchronous generators. Distributed gridplants' dynamic behaviour during fault transients was analyzed using an advanced machine learning (ML)model. Experiments conducted on the IEEE 118-bus system demonstrated the effectiveness of the method,achieving an accuracy of 95%, recall of 0.91, f1-score of 0.94, and precision of 0.92 which was higherthan the state-of-the-art models. Modern power grid automation relies on the suggested methods, whichimprove fault localization accuracy, strengthen system resilience, and provide practical isolationmeasures. While the proposed method contributes to improved localization, its primary innovation lies inaccurately localizing fault points within complex distributed networks using the FLSO-SVR framework.