Intelligent spatiotemporal-fused deep learning resilient framework for maritime cyber-physical systems

Abstract Modern shipboard power systems have increasingly integrated cyber-physical systems with the digital transformation of the maritime industry, thereby exposing them to unprecedented cybersecurity threats. Among these, false data injection attacks (FDIA) present a particularly severe challenge to vessel operational safety because they can evade conventional detection methods and subtly manipulate system state measurements while remaining concealed. To address this critical challenge, this study proposes an integrated framework referred to as CNN-BiLSTM-GAM, which is designed for effective FDIA detection and accurate data recovery in shipboard power systems. The framework synergistically integrates multiple deep learning architectures: a convolutional neural network (CNN) extracts spatial features from multivariate sensor data affected by attacks, a bidirectional long short-term memory (BiLSTM) network captures bidirectional temporal dependencies in power system operational data, and a global attention mechanism (GAM) dynamically weighs the spatiotemporal features, enabling the precise identification of subtle anomaly patterns induced by attacks in the highly dynamic and noisy environment of shipboard power systems. Compared with existing studies, experimental results demonstrate that the proposed CNN-BiLSTM-GAM framework achieves outstanding performance in attack detection, with an accuracy of 97.71% and an F 1-score of 0.9548. In data recovery, it achieves a coefficient of determination ( R 2 ) of 0.9686, significantly outperforming various baseline models.