AUTOMATIC FEATURES EXTRACTION BY TRANSFER LEARNING FOR TRANSMISSION LINE PROTECTION

Abstract

This work proposes a deep learning-based fault detection and classification model with relaxed dataset requirements. The most arduous part of any deep learning-based solution is the availability of large, labeled datasets. The proposed method uses a pre-trained deep learning model as a starting point, then retrains the adapted weight in transfer arrangement for fault classifier applications. This strategy expedites training and reduces the need for exhaustive labeled dataset requirements by leveraging an existing model. The proposed model automatically extracts features from input signals to decide the state of power transmission lines, eliminating the complex need to craft features for fault classification algorithms manually. The model is thoroughly tested for a wide range of performance tests. (The dataset used in this work is publicly available at this URL: https://www.kaggle.com/datasets/fezanrafique/wsccc9busfaultdataset).