Global warming has exacerbated flood disasters, leading to severe Natural hazard triggered technological accident(Natech). Chemical plants are particularly exposed because their process continuity, safety functions, and hazardous material storage rely extensively on electrical systems that are vulnerable during flooding.To address the lack of systematic analysis of flood-induced electrical failures, this study investigates flood-related Natech events and extracts equipment-level damage information using the large language model(LLM), enabling a structured analysis of heterogeneous accident reports.Based on 48 detailed cases, three categories of failure causes are identified and organized into a fault tree(FT), which is further mapped into a Bayesian network(BN) to overcome the limitations of traditional binary logic analysis. Probabilistic inference and sensitivity analysis are conducted to quantify causal relationships and assess the relative importance of contributing factors.The results indicate that flood intensity, terrain conditions, deficiencies in flood protection design, insufficient waterproofing of electrical equipment, and failure of redundant systems are the most critical contributors to electrical equipment damage during floods. A representative case study demonstrates how electrical power loss can escalate into severe consequences when temperature-sensitive chemicals are involved. Finally, resilience-oriented measures related to site layout, equipment design, waterproofing, and emergency preparedness are proposed to improve Natech risk prevention in chemical plants. • Utilizes the LLM to extract both structured and unstructured data. • Identified human, environmental, and mechanical factors of electrical failures. • Quantified causal pathways of flood-induced electrical damage through Bayesian network. • Highlighted critical factors for improving flood risk prevention of chemical plants.
Li et al. (Sun,) studied this question.