Cooperative Optimization Method Of Parameters And Structure Of Belief Rule Base And Application

Belief rule base(BRB)is a comprehensive complex system modeling method developed on the basis of Dempster Shafer evidence theory,decision theory and fuzzy theory.Researchers have conducted in-depth research on the optimization of parameters and structure of BRB model,but most of the existing studies consider the optimization of BRB's parameters and structure separately,and cannot achieve the collaborative optimization of the two.The model parameters and structure determine the accuracy and scale complexity of BRB modeling respectively.it is more necessary to consider them simultaneously in order to find the balance between accuracy and complexity.In view of this,the collaborative optimization method of parameters and structure of belief rule base and its application are studied in this paper:(1)BRB single objective optimization method based on parallel multi population and redundant gene strategy.Parallel multi population strategy is used to optimize different BRBs(with different number of rules)at the same time;redundant gene strategy is used to ensure that these BRBs can carry out optimization operations such as crossover / mutation;finally,the optimal individuals in each population are found according to the fitness of differential evolution algorithm,and these individuals constitute a BRB model with optimal parameters and structure(equal to the population's number).Because the selection of the optimal individuals in different populations is obtained by comparing the fitness of individuals in each population,the individuals of different populations do not interact with each other,which is equivalent to multiple tasks to select the optimal individual independently,so the method is essentially a single objective optimization method.(2)BRB multi-objective optimization method based on dominant subordinate framework.In order to further improve the accuracy of the model,the dominant subordination framework is introduced to generate and update the Pareto frontier by minimizing the mean square error of the system and the number of BRB rules;in multi thread parallel optimization mechanism is used to allocate different populations(the number of individuals in the population is the same)to multiple threads of subordination optimization,and multiple threads are parallel computing to improve the diversity of the population and prevent the solution of the model falling into local optimum.In the process of dominant optimization,the updating of Pareto frontier is obtained by comparing the fitness values of all individuals in different populations with the optimal solutions on the Pareto frontier(the number of solutions lower than the population),and the individuals of different populations can interact with each other,so this method is a multi-objective optimization method.(3)Verification and analysis of the proposed methods by the estimation of oil pipeline leakage.The BRB model is used to establish the nonlinear relationship between the input(pressure difference and flow difference)and the output(pipeline leakage value).The estimated value of the output can be derived by BRB model after the input sample value is obtained online.The single objective and multi-objective optimization methods proposed above are applied to optimize respectively BRB models and test the detection and estimation effect.It is concluded that the former is more than the latter in the number of solutions,and the latter is higher than the former in the estimation accuracy.In solving engineering problems,it can be selected and used according to the actual needs.