Deep Learning Based Overcurrent Relay Supervision Against Cyberattacks: The Role of Data Pre-Processing

Modern power system substations are digitalized and integrated with different communication technologies. As a result, the power systems became more susceptible to cyber threats. This paper details the application of a deep learning model called convolutional auto-encoder, and the importance of pre-processing the training data to discriminate actual fault from cyberattacked faults in overcurrent relays. The deep learning model is trained utilizing fault data that has undergone data pre-processing. It is discussed how data pre-processing techniques affect the effectiveness of the deep learning model. This study takes into account different cyberattack scenarios, including replay attacks, CT tap setting attacks, and false data injection attacks. The results from testing and validations shows that the suggested strategy can outperform the existing method.