Transmission Line Fault Detection and Classification Using Feature Extraction Based Self-Attention Convolutional Neural Network with Time Series Image

The transmission lines continuously experience the number of shunt faults and its effective in practical system increases the instability, load damages and line restoration cost. This work implements an advanced self-attention convolutional neural network (SAT-CNN) method for fault detection and classification (FDC) of high voltage transmission lines experiencing a constant number of shunt faults. Faults cause load damage in real-time applications, increase instability, and increase the cost of line restoration. Therefore, a precise model is needs to identify and categorize the flaws to quickly restore he problematic phases. In this research, implemented SAT-CNN feature extraction model with imaging-based on time series that can accurately detect and classify faults. By employing a number of input signals, including, current, voltage, and combined current and voltage signals, at different sampling frequencies, the efficacy of implemented SAT-CNN model is evaluated. Implemented SAT-CNN method obtains high performance when compared to existing methods including weight-sharing network (WSCN), Truncated singular value decomposition and Human urbanization algorithm based Recurrent Perceptron Neural Network (TSVD-HUARPNN), and Graph convolutional neural network (GCN), and result achieved 99.90% accuracy value.