28 A Method for Human Reliability Assessment Based on Bayesian Networks (PSAM-0443)

This article presents a Bayesian network model for human reliability assessment (HRA). In most existing HRA methods, the probability of human error is determined by a set of influencing factors. Each factor in the human reliability assessment is represented as a node in the Bayesian network. The proposed method can be applied to both qualitative and quantitative reliability analysis. For the qualitative analysis, the causalities of the nodes (the factors in HRA) are directly visualized by a hierarchical graph in which we can easily locate the factors that should be considered for improvement. As for the quantitative analysis, the posterior probability (the potential factor) is inferred by the prior information (including simulation data, field data and experts' knowledge) and the latest information on HRA. In this way, a certain potential human action can be predicted by calculating the node's posterior probability. We also suggest the use of Bayesian factor analysis to determine the confidence level of the model if the prior probability distribution and posterior probability distribution for each node are known. In addition, the Bayesian network is easy to be extended by adding new factors.