Use of evidential reasoning for eliciting Bayesian subjective probabilities in human reliability analysis

Abstract

Modelling the interdependencies among the nine common performance conditions (CPCs) in Cognitive Reliability Error Analysis Method (CREAM) stimulates the use of Bayesian Networks (BNs) in Human Reliability Analysis (HRA). However, subjective probability elicitation for a BN is often a daunting and complex task. To create conditional probability values for each given variable in a BN requires a high degree of knowledge and engineering effort, often from a group of domain experts. This paper presents a new hybrid approach for combining an ER algorithm with BNs to enable HRA under incomplete data. The kernel of this approach is to develop the best and the worst possible conditional degrees of belief of the nodes influencing Contextual Control Model Controlling Modes (COCOM-CMs) when using BNs to model human error quantification in CREAM. The findings on the hybrid evidential reasoning and BN model can effectively facilitate human failure probability analysis in CREAM in specific and decision making under uncertainty in general.