Fault Detection and Localization in Electrical Transmission Lines Using ELM and LSTM Networks Based on MATLAB/Simulink Modeling and Simulation

Abstract

Overhead transmission lines ensure continuous power delivery but remain vulnerable to environmental and operational disturbances, making faults a persistent threat to grid reliability. In MATLAB/Simulink, we simulate a dual-voltage network with five $20-\mathrm{km}$ lines, injecting multiple fault types at varied locations. We build a synthetic dataset of root-mean-square threephase voltages and currents from these runs and train a twostage framework: Long Short-Term Memory (LSTM) networks for detection/localization and Extreme Learning Machines (ELM) for fast inference. In a six-class task, the LSTM achieves 86% accuracy, demonstrating strong predictive capability; the much faster ELM reaches 61%, revealing a speed-accuracy trade-off. Results indicate intelligent, data-driven monitoring can shorten response times and enhance stability in high-voltage transmission systems. Future work targets better generalization, real-time stream integration, and deployment within operational monitoring platforms. The framework supports single- and multi-line fault cases, varying inception angles and resistances, and yields interpretable outputs suitable for operator decision support.