A Deep Neural Network Based Approach for Fault Detection and Localization in Power System Network

Abstract

Ensuring reliable operation in electricity transmission networks requires efficient fault detection, classification, and localization. However, the integration of distributed generators and the dynamic nature of these systems pose challenges to traditional relaying devices in managing fault currents. This study investigates the application of deep learning techniques to address these challenges by autonomously extracting fault characteristics from three-phase voltage and current signals. Using Artificial Neural Networks (ANNs) and one-dimensional Convolutional Neural Networks (1D-CNNs), fault detection, classification, and localization are performed on the IEEE 9-bus system. Simulated fault data is generated in MATLAB/Simulink, and deep learning models are trained using Python libraries such as scikit-learn and TensorFlow. Results indicate high accuracy, with 1D-CNN achieving 99.87% for faulty line identification, 92.42% for fault classification, and 96.95% for fault location. Similarly, the ANN model attained 99.54%, 92.35%, and 96.24%, respectively. To optimize the cost and complexity of phasor measurement unit (PMU) deployment, a selective feature reduction strategy was implemented, focusing on critical buses (5, 6, and 8), demonstrated that fault analysis can be effectively performed with reduced data inputs, while minimizing the required PMUs. Additionally, transfer learning for N-1 contingency scenarios allowed the pre-trained models to efficiently adapt to new cases, enhancing fault diagnosis performance. These findings highlight the potential of deep learning to improve the accuracy and reliability of fault diagnosis in power transmission systems, supporting future real-time implementation.