A Bayesian Network Inference Approach for Dynamic Risk Assessment Using Multisource-Based Information Fusion in an Interval Type-2 Fuzzy Set Environment

The Bayesian network (BN) method has been identified as a research hotspot in dynamic risk assessment (DRA) for systems. The traditional BN inference process relies on crisp probabilities; however, it is inapplicable in an interval type-2 fuzzy set (IT2FS) environment. This research aimed to fill this gap by developing a BN inference approach for DRA using multisource-based information fusion in an IT2FS environment via the following stages. In stage A, a fusion rule for the upper and lower membership degrees of multisource IT2FS-based information was defined using the fuzzy granulation method, and an information fusion algorithm in an IT2FS environment (Algorithm 1) was developed to fuse the multisource IT2FS-based information provided by experts. In stage B, the structure and conditional probability tables of the BN model were determined, and a BN model for conducting the DRA in an IT2FS environment was built using the BN method. In stage C, a cumulative distribution function of fused multisource IT2FS-based information was defined, the traditional Latin hypercube sampling (LHS) method was improved, and a novel BN inference algorithm for implementing the DRA in an IT2FS environment (Algorithm 2) was developed using the improved LHS method. The developed BN inference approach was applied in a representative case, and the application results showed that BN inference could effectively predict dynamic risk and analyze risk sensitivity.