Cluster-based information fusion for probabilistic risk analysis in complex projects under uncertainty

Abstract

This paper proposes a hybrid soft computing approach that integrates the Dempster–Shafer (D–S) evidence theory and cluster analysis for probabilistic risk analysis in complex projects under uncertainty. The fusion model tends to solve multi-criteria decision-making problems with a focus on the information content reflected from evidence. Risk factors are quantified into a continuous numeric scale for risk level classification and each factor value is turned into a basic probability assignment (BPA). A sorting operator is used to aggregate the evidence into risk level based clusters. The D–S evidence theory is first used to fuse similar evidence within each cluster, and then the weighted ratio method is used to fuse conflict evidence between clusters. The fused result is defuzzied into a crisp value to give a conveniently referred value for decision-making. Global sensitivity analysis is conducted to depict the effect of each risk factor on the overall estimated risk level. The developed approach is used to assess the water leakage condition of Line 2 of the Wuhan metro system in China to demo its feasibility. The tunnel is assessed to lie in a good condition with a tolerance of 5% measurement error. The proposed two-step fusion process is capable to reserve more details through computation and enhances the confidence in risk classification results compared to that based on a separate piece of evidence. This research contributes to (a) a systematic classification and fusion-based quantitative risk analysis method; (b) practical risk assessment of water leakage in operational tunnels.