Abstract
The increasing integration of digital technologies in hybrid hydrogen-power networks has introduced new cybersecurity vulnerabilities that existing static or single-phase cyberattack models fail to adequately exploit or defend against. These models typically lack dynamic adaptability, coordination across multiple attack stages, and obfuscation mechanisms, thereby limiting their effectiveness and realism. To address this gap, we propose a novel Cyberattack Design Based on CNN-DQN-Blockchain Technology for Targeted Adaptive Strategy (CDB-TAS)-a three-stage, dynamically evolving cyberattack framework tailored for hybrid hydrogen-electric networks. The proposed CDB-TAS model comprises: (i) a Preliminary Reconnaissance Phase, where a Convolutional Neural Network (CNN) identifies the most vulnerable buses via real-time anomaly detection; (ii) an Escalation Phase, where a Double Deep Q-Network (Double DQN) dynamically refines the attack strategy based on grid response and demand profiles; and (iii) a Sustained Attack Phase, which maintains high-intensity disruptions while minimizing detection through continuous feedback adaptation. Additionally, a private blockchain network is employed not for defense, but as an attacker-side obfuscation layer-concealing attack metadata and enabling decentralized coordination among malicious nodes. Simulations on a synthetic 2000-bus hybrid hydrogen-power system modeled after ERCOT reveal that CDB-TAS induces up to 15% voltage drop at critical buses (e.g., Bus 3103), disrupts over 600 MW of load across 50 substations, and achieves 23.4% higher disruption efficiency with lower anomaly detection rates compared to baseline attacks. This study presents the first integrated framework combining CNN, reinforcement learning, and blockchain from an adversarial perspective, offering new insights into the evolving threat landscape and guiding the development of future cyber-resilience strategies in multi-energy systems.