Deep Learning-based Fault Diagnosis in Transmission Lines via Long Short Term Memory Networks

In electrical power systems, transmission lines are responsible for transferring power across the grid. However, faults in these lines are abnormal conditions that can destabilize the transmission system if sustained longer. To diagnose the faults, IEC61850 based digital substations provide the sampled value measurements in the substation. In addition to existing model-based techniques to diagnose the fault, machine learning techniques are explored in the literature. In this thesis, we present a novel Long Short Term Memory (LSTM) based fault classifier using current and voltage measurements as the input. Compared with deep learning algorithms proposed in the literature i.e. RNN and SVM, the proposed classifier provide improved performance in the classification of faults, tested on data obtained from a PSRC D6 benchmark testbed. The performance of the classifier is explained with the evaluation metrics i.e. test accuracy, precision, recall, F1 score and confusion matrix to show the classification performance.M.A.S