基于主导从属框架的变结构置信规则库多目标优化方法

Multi-objective optimization method of variable-structured belief rule base using dominant subordinate framework

置信规则库(belief rule base,BRB)是一种十分有效的非线性建模工具,当前BRB优化相关研究仅局限于单目标优化.为了解决BRB的多目标优化问题,尤其需要解决在优化BRB参数的同时对BRB的结构也进行优化的问题,即BRB在优化过程中将变结构,本文提出基于主导从属框架结构(dominant-subordinate framework)的置信规则库多目标(multi-objective)优化方法(DSM-BRB).在主导优化过程中,将采用多种群策略和冗余基因策略对不同规模的BRB进行协同优化,其中多种群策略确保每个种群中的个体长度相等,不同种群中的个体长度不等,冗余基因策略将冗余基因添加到基因数量较少的个体中以确保每个个体具有相同的长度,因此不同种群中的个体可以协同优化,并在进入适应度计算之前删除添加的冗余基因.在主导优化过程完成之后,采用多线程并行优化机制将不同种群分配至多个从属优化线程中,即在一个线程中仅对一个种群中的个体进行优化,多个线程同时进行并行优化以提高优化效率.最后本文以某石油管道泄漏检测问题作为示例验证本文提出方法的有效性.

Belief rule base is a nonlinear modeling tool,but current optimization studies of belief rule base only are only limited to single-objective optimization.In this study,a multi-objective optimization approach of BRB is proposed where the parameters and the structure of BRB are optimized at the same time,specifically that the structure of BRB is optimized during the optimization process.In the proposed approach,a dominant-subordinate framework and two strategies are adopted,namely the multi-population strategy and redundant gene strategy.The multi-population strategy ensures that individuals in each population are of equal length,individuals in different populations vary in length.The redundant gen strategy denotes that every individual with insufficient gens are added redundant genes in order to ensure every individual have the same length so that individuals in different populations can be collaboratively optimized simultaneously.Before entering fitness calculation,BRB should delete redundant genes.The two strategies are both implemented in the dominant process.After completing the dominant process,different populations are allocated in different threads where each thread only hosts one population,which is the subordinate optimization process being implemented in a parallel fashion.A practical benchmark case of pipeline leak detection problem is studied to validate the effectiveness of the proposed method.