基于决策空间分段预测的动态多目标优化策略

打开文本图片集

中图分类号：TP273 文献标志码：A

Dynamic Multi-objective Optimization Strategy Based on Segmented Prediction of Decision Space

LI Erchao， LIU Chenmiao

（School of AutomationandElectrical Engineering，Lanzhou UniversityofTechnology，Lanzhou73Oo5O，Gansu，China）

Abstract：Toquicklyand accurately track Pareto solutionsetin a new environment，solve the problemsthatthe traditional single center point prediction was notaccurate and the numberof representative elite individuals was small，a dynamic multi-objective optimization strategy based onsegmented prediction of decision space （SPDS）wasproposed.Firstly， Pareto solution setobtained atthe previous moment was evenlydividedinto thre segments according to Euclideandistance to ensure the breadth ofthesearch space and search eficiency.Secondly，thecentral point moving stepof each segment Pareto solution set was obtained.Finall，the linear prediction mechanism was used to predict the next generation populationpiecewise，which madethealgorithmmorerobustandadaptable.Inordertoverifytheefectivenessof SPDS， dynamic non-dominated sorting genetic algorithm-II（DNSGA-II）-A，population prediction strategy（PPS）and prediction strategy basedonspecial point（SPPS）algorithms was compared withand without SPDSon15 standard dynamic test functions.SPDS-DNSGA-I-Aalgorithmwas appied totheoptimizationof proportional-integral-derivative parameters of dieselengines.Theresultsshow thattheoptimalrateof SPDSstrategyin termsof invertedgenerationaldistanceis 78.33% higher than those of the comparison algorithms，respectively. The convergence and diversity of SPDS strategy are improved to different degrees，which canadapt to dynamicenvironmental changesandefectivelysolve dynamic multi-objective optimization problems.

Keywords：dynamic multi-objective optimization；evolutionary algorithm；segmented prediction；decision space

现实生活存在大量2个或多个目标相互冲突的问题，同时还伴随有某种约束条件，该类问题被统称为多目标优化问题（MOP）[1]。（剩余15461字）