Detection and Classification of Transmission Line Faults Using Convolutional Neural Networks

Detecting and classifying faults in transmission lines are essential for ensuring power system stability and reliability. This paper introduces a standard method utilizing Convolutional Neural Networks (CNNs) to improve the precision of fault categorization. A 220 kV, 200 km transmission line model is used to simulate various fault scenarios, including single line-to-ground faults (AG, BG, CG), line-to-line faults (AB, BC, CA), double line-to-ground faults (ABG, BCG, CAG), and three-phase faults (ABCG). Voltage signals generated under these fault conditions are encoded into two-dimensional Gramian Angular Summation Field (GASF) images. A custom CNN architecture is developed, incorporating convolutional layers, pooling layers, and fully connected layers with ReLU and Softmax activation functions. Additionally, several pre-trained CNN models are modified using transfer learning and fine-tuning techniques to classify faults. The proposed CNN model attains an exceptional accuracy of 90-93%, surpassing pre-trained models such as ResNet, ConvNext, EfficientNet, etc. This study emphasizes the benefits of employing image-based representations of time-series data and illustrates the significance of deep learning for reliable transmission line fault detection across varying parameters. The comparative analysis with pre-trained architectures further substantiates the proposed model’s reliability and importance.