Improved fuzzy Bayesian network-based risk analysis with interval-valued fuzzy sets and D-S evidence theory

• Main Authors: Pan, Yue, Zhang, Limao, Li, ZhiWu, Ding, Lieyun
• Other Authors: School of Civil and Environmental Engineering
• Format: Article
• Language: English
• Published: 2022
• Subjects: Engineering::Civil engineering; Fuzzy Bayesian Network; Risk Analysis
• Online Access: https://hdl.handle.net/10356/161031
• Institution: Nanyang Technological University

A novel risk analysis approach is developed by merging interval-valued fuzzy sets (IVFSs), improved Dempster-Shafer (D-S) evidence theory, and fuzzy Bayesian networks (BNs), acting as a systematic decision support approach for safety insurance for the entire life cycle of a complex system under uncertainty. Aiming to alleviate the problem of insufficient and imprecise data collected from the complicated environment, the expert judgment in linguistic expressions is employed to describe the risk levels for all risk factors, which are represented by IVFSs using Gaussian membership function to fully consider such fuzziness and uncertainty. In regard to interval fusion and highly conflicting data, an improved combination rule based on the D-S evidence theory is developed. Then, fuzzy prior probability for each risk factor can be generated from fused intervals and fed into a fuzzy BN model for fuzzy-based Bayesian inference, including predictive, sensitivity, and diagnosis analysis. Furthermore, a case study is used to demonstrate the feasibility of the proposed risk analysis. A comparison of risk analysis based upon the hybrid improved D-S, classical D-S, and arithmetic average method is illustrated to show the outstanding performance of the developed approach in fusing multisource information with ubiquitous uncertainty and conflicts in an efficient manner, leading to more reliable risk evaluation. It is concluded that the proposed risk analysis provides a deep insight on risk control, especially for complex project environment, which enables to not only reduce the likelihood of failure ahead of time but also mitigate risk magnitudes to some degree after the occurrence of a failure.