Abstract: In order to accurately predict falling possibility at height in subway station construction, an evaluation model of it based on HFCAS-BN was constructed. Firstly, based on HFACS framework, causes of falling at height were identified, and the framework was mapped into BN. Secondly, group decision-making method based on normal distribution weight, fuzzy-BWM and global relative value were incorporated into BN to obtain prior probability of root nodes and conditional probability of intermediate nodes. Then, probability of falling nodes in various states was predicted and key causes were identified by using BN forward reasoning and sensitivity analysis. Finally, three stations of Chengdu Subway Line 11 were selected for case analysis. The results show that evaluations of most possible state of Huilonglu Stop, Diaoyuzui Stop and Lujiaocun Stop, respectively severe fault, fault free and fault free, are basically consistent with actual construction situation. Key causes of falling accidents include inadequate safety education and training, removal of falling prevention measures, losing one's footing, poor safety awareness, standing on unsafe places and inadequate lighting, so major efforts should be made on them.

Key words: human factors analysis and classification system (HFACS), Bayesian network (BN), subway station construction, falling at height, evaluation of possibility