An improved method for building conditional probability tables of Bayesian networks in human reliability analysis and its application for railway

Abstract

This paper presents an improved conditional probability tables (CPT) building method for Bayesian networks (BN), which is suitable for human reliability analysis (HRA). This method is based on an existing CPT assignment method in the field of HRA and the improvement is realized by using the fuzzy inference theory. Mamdani method is applied to replace parameter R assignment process in the original method. The experts only need to estimate the weight value of all the parent nodes of a child node in the BN, and then the CPT can be obtained through fuzzy reasoning. Compared with the original method, the workload of experts can be significantly reduced. The improved method also has better stability because it can greatly reduce experts' bias. In this paper, the improved CPT assignment method is applied to automatically assign the CPT of the BN in the HRA model of the train driver of high-speed railway. After quantitative analysis, the human reliability of the train driver can be calculated accurately.