A Hybrid Genetic Algorithm-Optimized Decision Tree Framework for Power System Fault Classification Using Symmetrical Component Analysis

Abstract

The growing complexity of modern power networks, driven by increasing load demand and intermittent renewable energy integration, has made traditional fault detection techniques less reliable and slower in responding to diverse and unbalanced fault scenarios. These challenges often result in delayed isolation of faulty sections, equipment stress, and reduced system reliability. This study presents an optimisation-based machine learning framework that integrates Decision Trees and Genetic Algorithms for accurate classification of power system faults under varying operating conditions. Symmetrical component transformation was used to extract meaningful sequence features from three-phase voltage and current signals. Fault currents for different scenarios, including single line-to-ground, line-to-line, and three-phase bolted faults, are analytically modeled using Thevenin and sequence network approaches. These features are then used to train a Decision Tree classifier, while a Genetic Algorithm optimizes its parameters by maximizing a fitness function that combines accuracy, precision, and computational speed. Simulation results demonstrate strong performance of the framework. The positive sequence current dominated at approximately 10.0 p.u., while the zero-sequence current remained around 1.0 p.u., validating the effectiveness of symmetrical component analysis. The single line-to-ground fault current reached a peak of 2.5 p.u. at a zero-sequence impedance of 0.1 p.u., and the three-phase bolted fault current attained 2.0 p.u. when the Thevenin impedance was 0.5 p.u. The classifier achieved an accuracy of about 95% when trained with 90% of the dataset, while the RMSE converged from 0.16 to a stable value near 0.06, indicating high prediction reliability. The Genetic Algorithm showed fast convergence, and the PSO-enhanced optimization ensured stable parameter tuning. From a policy perspective, this research supports the adoption of intelligent protection systems in national grids, encouraging utilities to integrate optimization-based machine learning tools for faster fault diagnosis, improved grid resilience, reduced outage duration, and enhanced operational safety in smart power systems.