A Novel Hybrid Signal Processing Based Deep Learning Method for Cyber-Physical Resilient Harbor Integrated Shipboard Microgrids

Abstract

Recently, the cybersecurity of harbor-integrated shipboard microgrids (SMGs) has become a significant concern for ensuring the integrity and resilience of maritime energy systems within intelligent grids. However, the increasing reliance on internet-based technologies exposes SMGs to false data injection attacks (FDIAs), denial of service (DoS) attacks, and other cyber threats, which can compromise power management and operational stability. To address these threats, this paper proposes a hybrid detection framework that integrates Hilbert–Huang Transform (HHT) for signal decomposition and anomaly identification, Long Short-Term Memory Variational Autoencoder (LSTM-VAE) for deep feature extraction and latent attack pattern detection, and a Bidirectional Gated Recurrent Unit (Bi-GRU) for sequential dependency analysis and real-time FDIA detection. The HHTbased empirical mode decomposition (EMD) enables timefrequency analysis by extracting intrinsic mode functions (IMFs), improving the detection of nonstationary attack patterns in maritime environments. The LSTM-VAE identifies subtle attack signatures within SMG data, while the Bi-GRU enhances temporal modeling and classification, ensuring accurate and real-time detection with minimal false positives. Extensive validation against various cyberattack scenarios, including component-level intrusions, demonstrates the framework's robustness in detecting and mitigating cyber threats. By optimizing the deep LSTM-VAE-based Bi-GRU model with carefully tuned hyperparameters, the proposed method achieves a detection accuracy of 99.80%, surpassing existing techniques. This work introduces a scalable, adaptive, and maritime-specific cybersecurity solution, significantly enhancing shipboard microgrid security and operational resilience in intelligent maritime SMGs.