A new belief rule base model with uncertainty parameters

In the current researches on belief rule base (BRB), input parameters are often represented by quantitative values, which limits the ability of BRB to resolve uncertainties. Obviously, uncertainty parameters are more suitable for modeling BRB, so a new BRB-UP model with uncertainty parameters is developed in this paper. In BRB-UP, attribute weight is described as interval value and activation weight is described as random variable. First, the analysis focuses on the mapping relationship between interval attribute weight and activation weight, and based on monotonicity of multivariate function, extreme value of the activation weight can be derived. Second, in the extreme range, activation weight is described as random variable, a new inference engine based on evidential reasoning algorithm (ERA) is proposed, and basic properties of the engine are proved. Third, considering differences in the number of activated rules, a double inference engine for BRB-UP is proposed. Finally, extensive experiments conducted on a JRC-7 M aerospace relay and the NASA lithium battery public datasets demonstrate that, although BRB-UP has a higher time complexity, it exhibits greater precision and stronger robustness compared to BRB.