Security Enhancement of Cyber-Physical DC Ship Power System Using Scalable Deep Learning Method

Cyber-physical systems like ship power systems are highly susceptible to cyber threats and physical faults. Early detection and mitigation of cyber-attacks using Machine Learning (ML) techniques are garnering more attention as cyber-physical attacks become more frequent and sophisticated. ML has demonstrated promising performances in developing robust threat intelligence solutions. In this paper, a cyber-attack detection scheme for the ship control system using a scalable federated deep learning-based method is proposed. Here, Long Short-Term Memory (LSTM) models are trained in a distributed manner to mirror complex controller behavior using normal system operational data. Estimation errors between the actual controller data and the ML counterparts are used to detect the presence of cyberattacks or anomalous behavior. Additionally, temporal analysis is conducted on both the input and output estimation error signals whereby a time gap is detected that is leveraged to distinguish between hijacking and FDI attacks. Simulation results indicate that the proposed methods can effectively detect multiple types of cyberattacks instantly without sharing sensitive data.