Research And Application Of Inference Method Of Vector-based Belief Rule Base

The Belief Rule Base(BRB)is a comprehensive nonlinear system modeling method that combines expert system,fuzzy set theory and Dempster-Shafer(D-S)evidence theory.The distributed belief structure is introduced to expressed the consequet attributes of the IF-THEN rule whitch can be used to process fuzzy uncertainty,uncertain probability even incomplete information.However,when there are many antecedent attributes(external variables)in questions,the "combination explosion" problem of rules often occurs,which affects the accuracy and applicability of the BRB methodology.To solve this problem,a vector-based belief rule base inference method is proposed and applied to engineering system abnormal detection and slope stability evaluation.The main contents are as follows:(1)A nonlinear causality function fitting method based on vector-based belief rule base inference.When using the inference method of the belief rule base,a nonlinear causal relationship model between external monitoring variables and internal state variables is usually established.To solve the "combination explosion" problem,multiple external variables are sorted according to their contribution rate.Then,vector-based belief rules are combined.The idea of space vector matching and "full activation" is introduced to improve the original matching and activation methods of the belief rule base.Thus,this method not only reduces the number of rules,but also ensures the accuracy of the fitting.Finally,the nonlinear function fitting experiment is used to prove the effectiveness of the method.(2)Pipeline leak detection method based on vector-based belief rule base inference.Based on the vector-based belief rule base(V-BRB)method proposed in content(1),the nonlinear mapping relationship among the oil pipeline input/output flow difference,pipeline pressure(other external variables)and the leakage is established.When new sample value is obtained online,it can be brought into V-BRB to infer the value of leakage.In addition,the sequential linear programming(SLP)method and historical data are used to optimize the parameters of the V-BRB model,which improves the accuracy of the V-BRB model.Finally,the oil pipeline leak detection experiment is used to verify the effectiveness of the method.(3)Slope stability evaluation method via belief rule base with vector-based high-dimensional mixed inputs.In the evaluation of slope stability,external variables(hazard factors)usually have two different forms: discreteness and continuation.Therefore,a discretization coding method is proposed to realize unified discretization and reduction of the two types of variables.Thereby the BRB model is established.Then the discrete/continuous external variables obtained online can be brought into the BRB model to infer the stability grade of the slope.This method reduces the scale of the BRB and ensures the accuracy of the slope stability evaluation grades.Finally,the effectiveness of this method is verified through the evaluation experiment of slope stability in the mining area.