摘要
研究表明,现有的多目标进化算法在处理具有不同Pareto前沿的优化问题时难以有效平衡种群的收敛性与多样性.鉴于此,提出一种基于自适应参考向量和参考点的高维多目标进化算法(adaptive reference vector and reference point based many-objective evlolutionary algorithm, ARVRPMEA). ARVRPMEA主要利用种群稀疏性自适应调整参考向量和参考点以提高种群多样性,首先,生成均匀分布的参考向量子集和参考点子集,并利用该参考向量子集分解种群;然后,根据规模最大子种群中解的分布情况生成新的参考向量和参考点,直至满足参考向量集和参考点集规模;最后,为进一步提高种群收敛性,该算法结合指标进行环境选择以保存收敛性较高的个体进入下一代种群.实验结果表明, ARVRP算法在求解具有不同Pareto前沿的问题方面具有良好的性能.
The research shows that the existing multi-objective evolutionary algorithms are difficult to effectively balance the convergence and diversity of the population when dealing with optimization problems with different Pareto fronts.To address the above situation,this paper proposes an adaptive reference vector and reference point based many-objective evlolutionary algorithm(ARVRPMEA),which mainly uses population sparsity to adaptively adjust reference vectors and reference points to improve population diversity.First,the ARVRPMEA generates a uniformly distributed subset of reference vectors and a subset of reference points,and uses this subset of reference vectors to decompose the population.Then,new reference vectors and reference points are generated according to the distribution of solutions in the largest subpopulation until the scale of the reference vector set and reference point set is satisfied.Finally,to further improve population convergence,the algorithm combines the metrics for environmental selection to preserve the individuals with higher convergence into the next generation of populations.The experimental results show that the ARVRPMEA has good performance in solving problems with different Pareto fronts.
作者
覃灏
李军华
黎明
徐三水
QIN Hao;LI Jun-hua;LI Ming;XU San-shui(Key Laboratory of Jiangxi Province for Image Processing and Pattern Recognition,Nanchang Hangkong University,Nanchang 330063,China)
出处
《控制与决策》
EI
CSCD
北大核心
2024年第3期759-767,共9页
Control and Decision
基金
国家自然科学基金项目(62066031,61866025,61866026)
南昌航空大学研究生创新基金项目(YC2020-030)。
关键词
进化算法
高维多目标进化算法
自适应参考向量和参考点
分解种群
evolutionary algorithms
many-objective evolutionary algorithm
adaptive reference vector and reference point
decomposing population
