Abstract: To explore the intrinsic relationships among the contributing factors of gas leakage accidents and identify the key and root causes, this study collected 148 gas leakage accident investigation reports as textual data. The Latent Dirichlet Allocation (LDA) model was applied for topic mining, and 18 accident causative factors were identified and summarized through literature analysis. An integrated approach combining the Decision-Making Trial and Evaluation Laboratory (DEMATEL) and Interpretive Structural Modeling (ISM) was adopted to plot cause-and-result diagrams of the factors and construct a multi-level hierarchical structure model. This allowed for determining the causal attributes of each factor and their roles within the system, as well as clarifying the hierarchical relationships and transmission paths among the factors. The results show that the causal factors of gas leakage accidents exhibit a five-level hierarchical structure. Among them, inadequate safety supervision and insufficient implementation of safety education and training are identified as root causes, while violations of operating procedures and failures of auxiliary equipment and facilities are the most direct surface-level causes. A total of 11 key causal chains from root causes to surface-level factors were identified, revealing the multi-level transmission paths among the causes of gas leakage accidents and providing clear directions for accident prevention and control.

Key words: gas leakage accident, causative factors, LDA model, DEMATEL-ISM method, risk analysis