考虑帕累托最优解的多目标优化进化算法  

Multi-objective Optimization Evolutionary Algorithm Considering Pareto Optimal Solution

作　　者：王松波[1] WANG Song-bo(Maoming Polytecnic,Maoming 525000,China)

机构地区：[1]茂名职业技术学院,广东茂名525000

出　　处：《数学的实践与认识》2022年第9期132-146,共15页Mathematics in Practice and Theory

基　　金：茂名市科技专项资金项目:《基于微信小程序的O2O同城货运平台研发》。

摘　　要：大多数现有的进化算法在处理多目标优化问题(multi-objective optimization problem,MOP)时会遇到Pareto最优解稀疏的困难,特别是当决策变量的数目很大时,如旨在从大量候选特征中找出小部分特征的特征选择.为此,提出了一种求解大规模稀疏MOP的进化算法.算法考虑Pareto最优解的稀疏性,提出了一种新的种群初始化策略和遗传算子,以保证解的稀疏性.此外,还设计了一个测试套件来评估该算法在大规模稀疏MOP中的性能,实验结果和应用实例证明了该算法在处理大规模稀疏MOP问题上的优越性.Most of the existing evolutionary algorithms encounter the difficulty of Pareto optimal solution sparsity when dealing with multi-objective optimization problem(MOP),especially when the number of decision variables is large,such as feature selection aiming at finding a small part of features from a large number of candidate features.In this paper,an evolutionary algorithm for large-scale sparse mop is proposed.Considering the sparsity of Pareto optimal solution,a new population initialization strategy and genetic operator are proposed to ensure the sparsity of solution.In addition,a test suite is designed to evaluate the performance of the algorithm in large-scale sparse mop.Experimental results and application examples show that the algorithm has advantages in dealing with large-scale sparse mop.

关 键 词：进化神经网络 特征选择 多目标优化 稀疏Pareto最优解 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]