Artificial-intelligence-driven wide-area fault location for transmission systems: A traveling wave-based approach

Accurate and timely fault location (FL) is essential for improving grid resilience, minimizing service interruptions, and accelerating system recovery in power transmission networks. Among the various approaches, traveling wave (TW)-based methods-particularly wide-area TW-based techniques-offer high speed and accuracy, making them attractive for modern grids. However, existing TW-based methods often rely on ideal assumptions such as uniform propagation velocities, perfect sensor coverage, and noise-free measurements. These limitations hinder their performance in realistic settings, where line parameters vary, sensors may fail, and data may be incomplete or distorted. This paper presents a novel approach that leverages TW arrival times and a multi-task artificial neural network (ANN) architecture to jointly identify the faulty line and the FL. By formulating sensor placement as an optimization problem and solving it via Integer Linear Programming (ILP), the method significantly reduces hardware requirements while preserving high accuracy. Extensive simulations on a modified IEEE 39-bus system—featuring realistic grid conditions including underground cables, overhead lines, hybrid lines, and substation impedance—show that, for the final enhanced configuration, the proposed method achieves a classification accuracy of 99.62% and a mean absolute fault-location error (MAE) of 166.16 m.