Belief-Rule-Base Inference Method Based on Gradient Descent With Momentum

The belief-rule-base (BRB) inference methodology using the evidential reasoning (ER) approach is widely used in different fields, such as fault diagnosis, system identification, and decision analysis. However, calculation characteristic of conventional rule activation weight makes have zero problem. difficulty constructing partial derivatives restricts optimization parameters gradient method. Hence, this paper proposes a new belief structure its training method to solve problem during process improve accuracy. Gaussian function applied calculate with structure. Its characteristics avoid caused by attribute reference value set original Based on newly proposed method, corresponding distance-sensitive parameter for each attribute, discarded. It simplifies weights enables be easily constructed. In optimization, momentum stochastic descent train BRB system, which improves speed accuracy compared methods. Experiments nonlinear fitting, oil pipeline leak detection, classification several public datasets are carried out verify whether trained (SGDM-BRB) has better performance than other experimental results show that case complete data, SGDM-BRB higher faster