Integrating Large Language Models with Systemic Methods: A Hybrid Approach for Resilience Modeling of Hydrogen Refueling Stations

Hydrogen refueling stations (HRS) are critical components of the hydrogen energy infrastructure yet face significant safety challenges. Existing risk assessment methods, although widely used, often depend on subjective knowledge and struggle to capture dynamic interdependencies in complex operational scenarios. To address these limitations, this study proposes a hybrid resilience assessment methodology integrating large language models (LLMs), functional resonance analysis method (FRAM), Bayesian networks (BN), and complex network (CN) theory. The method leverages a tailored GPT-4 to automatically generate and refine functional nodes based on a domain-specific accident database, constructing a comprehensive FRAM model of HRS operations. Functional couplings and resilience roles are then identified and mapped to a BN structure, with node importance quantified by a k-shell entropy weight algorithm. A case study demonstrates the applicability and effectiveness of the proposed method, enabling a systemic approach to identifying and mitigating functional resonance, thereby ensuring the operational continuity and safety of HRS under complex perturbations.