ABSTRACT This study presents a comprehensive analysis of system resilience and recovery in the face of cyber attacks on a hydrogen energy infrastructure, utilizing advanced modelling techniques and state‐of‐the‐art cybersecurity strategies. We developed a robust optimization framework integrated with graph neural networks (GNNs) to detect and mitigate sophisticated cyber threats. The GNNs were trained on a dataset that included both normal operational data and simulated attack scenarios, enabling them to identify subtle patterns indicative of intrusions. Once an attack was detected, the system employed a stochastic dual dynamic programming (SDDP) approach to reconfigure operations dynamically, optimizing system performance while minimizing disruption. This dual‐layer defence mechanism–comprising initial detection by the GNN and subsequent mitigation through robust optimization–was tested against various cyber threat scenarios, including data injection, denial of service (DoS) and control system hijacking. Our findings reveal that the automated adaptive recovery (AAR) Strategy, which integrates real‐time monitoring and AI‐driven adaptive response, significantly outperforms traditional methods. Specifically, the AAR Strategy restored system performance to 90 percent within 60 minutes post‐attack, compared to only 70 percent recovery under conventional approaches. A 3D surface plot analysis further demonstrated that system performance declines sharply under prolonged high‐load conditions, with potential performance drops to below 20 percent when the load exceeds 80 percent over a 100‐min period. These results underscore the critical need for integrating adaptive and automated resilience strategies, like the AAR Strategy, into energy infrastructures. Our research contributes to the optimization of cybersecurity measures, offering a robust foundation for future advancements in the resilience of critical energy systems against evolving cyber threats.
Hua et al. (Thu,) studied this question.