利用冲突信息降维的进化高维目标优化算法  

Objective Reduction Using Conflict Information in Many-objective Optimization Evolutionary Algorithm

作　　者：罗乃丽[1,2] 李霞[1,2] 王娜[1,2] 

机构地区：[1]深圳大学信息工程学院,广东深圳518060 [2]深圳市现代通信与信息处理重点实验室,广东深圳518060

出　　处：《信号处理》2017年第9期1169-1178,共10页Journal of Signal Processing

基　　金：深圳市科技计划项目(JCYJ20160422112909302;GJHS20160328145558586)

摘　　要：进化多目标优化算法求解高维目标优化问题面临收敛能力、计算复杂度、决策以及Pareto前沿的可视化等困难,其根本原因是目标空间维数高。目标降维通过丢弃冗余目标,为缓解高维目标优化求解困难提供一种新思路。本文提出利用冲突信息降维的分解进化高维目标优化算法(CIOR-MOEA/D)。该方法通过衡量目标在近似解集上体现的冲突性,构造问题的冲突信息矩阵,对该矩阵进行特征分析,确定目标的重要性程度,实现维数约简,并利用分解进化多目标优化算法(MOEA/D)对重要子目标集合进行分解进化,从而得到问题的近似解集。实验结果表明,本文提出的目标降维算法在降维的准确性与鲁棒性上均表现突出,能够有效地处理冗余高维目标优化问题。It is well-known that multi-objective evolutionary algorithms may suffer from convergence, computational com- plexity, decision-making and visualization of the Pareto fronts while solving many-objective optimization problems with high dimensionality of the objective space. Objective reduction provides an alternative for many-objective optimization by discar- ding the redundant objectives. This paper proposes objective reduction using conflict information in many-objective evolu- tionary algorithm which is based on decomposition, short for CIOR-MOEA/D. It measures the conflict information among objectives in the approximation solution set. By constructing conflict information matrix of the original problem and identif- ying the importance of objectives via matrix eigen-decomposition, dimension reduction of the original problem is achieved. Lastly, MOEA/D is used to optimize the important objectives, and the approximate solution set is obtained. Experimental results show that the new conflict information based objective reduction is outstanding in accuracy and robustness, and it can effectively solve the reducible MaOPs.

关 键 词：多目标进化算法 高维目标优化问题 目标降维 冲突信息 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程] TP183[自动化与计算机技术—控制科学与工程]