A new paradigm for adaptive cyber-resilience of DC shipboard microgrids using hybrid signal processing with deep learning method

Abstract

Integrating electrification and digitization within dc shipboard microgrids (SMGs) has led to transformative changes and unprecedented advancements in the maritime industry. This transition has exposed vulnerabilities, particularly in distributed generation units (DGUs) and power converter configurations, necessitating robust defense against potential cyber threats that have the potential to cause system instability and, in extreme situations, result in the blackout of dc SMGs. This article proposes a new paradigm for adaptive cyber-resilience of dc SMGs using hybrid signal processing with deep learning (DL) methods to maintain the system’s resilient operation. Signal-processing techniques incorporating wavelet transform (WT) and singular value decomposition (SVD) have been developed to facilitate a thorough analysis of power converter configurations for early detection and mitigation of cyber threats. A deep 1-D convolutional neural network (1D-CNN) with the Adam optimizer based on SVD is then used for signal feature extraction. Multiple-input basis models for 1D-CNN have also been developed to automatically capture wavelet singular values from the raw fluctuation signals. The 1D-CNN-based autoencoder–decoder framework ensures diverse basis patterns, and the precision-driven 1D-CNN model weighting strategy optimizes the ensemble for attack detection. Test results of a typical dc SMG have shown the efficiency and reliability of the proposed method in achieving a higher accuracy score of 95.75% compared to other state-of-the-art techniques in attack detection across diverse scenarios in dc SMGs.