ANN-Based Fault Classification in Transmission Lines: Comparative Study of Activation Functions

Ensuring the reliability of power transmission systems requires efficient fault detection mechanisms. This study delves into the usage of Artificial Neural Networks (ANN) for classifying faults in transmission lines, comparing the performance of three activation functions—Tanh, Sigmoid, and ReLU. Detecting faults promptly with great accuracy is important for maintaining the stability and efficiency of power systems. Traditional methods, including wavelet transforms and conventional ANN models, have been widely used but often face challenges in achieving optimal accuracy and minimizing computational losses. In this study, a Multi-Layer Perceptron (MLP)-based ANN is trained on real-time transmission line data, utilizing voltage and current measurements from all three phases. The model is evaluated using classification accuracy and loss functions to identify the optimal activation function for reliable fault classification. Performance metrics, including model loss and convergence behavior, are analyzed to identify the optimal activation function for enhanced reliability. It is very evident from the results, the potency of ANN in fault classification while also showing us the importance of activation functions. This study offers a comparative evaluation to guide the selection of activation functions for improved transmission line fault detection in modern power grids.