Deep Learning-Based Attack Detection Against Smart Power Distribution System

The increasing digitalization and interconnectedness of power systems have rendered cybersecurity a critical concern. This paper presents a deep learning-based approach for detecting cyber-attacks on overcurrent protective relays in power distribution systems, leveraging LSTM (Long Short-Term Memory) autoencoders. The proposed system utilizes three-phase current datasets obtained from the IEEE 13-node system simulations to identify anomalies indicative of intrusions, particularly in the context of false data injection attacks (FDIA). The obtained results demonstrate that the LSTM autoencoder effectively differentiates between normal operational data and FDIA signatures. This study highlights the potential of deep learning in enhancing the resilience of power systems against emerging cyber threats, paving the way for more secure and reliable operations in modernized distribution grids. The investigation into the impact of various targeted attack phases demonstrates that the approach performs good for all the possible attacked phases and their combinations.