Domain-specific large language model-driven risk analysis of battery energy storage systems

Battery energy storage systems (BESS) are increasingly widely-used in industrial and civil fields. With the occurrence of related accidents, the safety of BESS is becoming pyramidally prominent. Risk analysis is essential for implementing risk prevention and control measures of BESS. However, due to data deficiency, conventional risk analysis methods heavily rely on subjectivity-prone expert knowledge. In this paper, aiming at decreasing the reliance of risk analysis on subjective information, we propose a risk analysis method integrating the domain-specific large language model (LLM), functional resonance analysis method (FRAM), and Bayesian networks (BN). Generated from domain-specific LLM, FRAM is subsequently mapped into a BN for quantitative risk analysis. The proposed method is validated through recent worldwide BESS accidents. Our results provide valuable references for the risk quantification and management of BESS.