A safety assessment model based on belief rule base with new optimization method

Safety assessment is an important aspect of health management for complex system. Belief rule base (BRB) is one of the expert systems which can handle uncertainty, ambiguity and conflicting information. In safety assessment based on BRB, its initial parameters are determined by experts and then modified by optimization models. In current studies, some intelligent optimization algorithms are applied, and the parameters are trained based on the generated random population. The optimized parameters and structure of BRB by these optimization models may lose physical meaning, and it loses interpretability. Thus, to ensure the modeling transparency and traceability, a safety assessment model based BRB with a new optimization method based on the method of feasible direction (MFD) is developed for the first time, where the gradient of output to model parameters is deduced. Moreover, the convergence of the optimization method is proved to ensure that the optimized parameters are optimal solutions. In the new optimization model, the parameters are trained based on the output gradient of BRB analytical model that can keep the transparency of modeling process and ensure the interpretability of the constructed safety assessment model. A case study is conducted to illustrate the effect of the developed safety assessment model.