Scholarlink

<p>In dynamic multiobjective optimization problems (DMOPs), quickly tracking the Pareto fronts under various changing patterns has been a key challenge. Prediction driven methods have shown strong potential in capturing these changing patterns; however, most existing approaches rely on linear models, which limits their effectiveness in addressing nonlinear changes. To address this issue, we propose a novel dynamic multiobjective evolutionary algorithm named LDFP-DMOEA that combines a linear strategy with a diffusion model. Specifically, the population is adaptively divided into two subpopulations according to the correlation of individual variations across consecutive environments. A linear prediction strategy is applied to the highly correlated subpopulation to exploit stable trends, while a diffusion model is employed for the less correlated subpopulation to capture complex nonlinear dynamics. The two subpopulations are then merged, followed by a local adjustment mechanism to enhance diversity. Experimental results show that LDFP outperforms five representative comparative algorithms on multiple benchmark problems, validating its effectiveness and superiority.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

A hybrid prediction based dynamic multiobjective optimization algorithm via a linear strategy and diffusion model

  • 期刊论文

  • 发表日期：

    2026年7月29日

查看全文
  • Yuting Du,
  • Lei Wang,
  • Qiaoyong Jiang,
  • Lanxing Zhang,
  • JiaJie Wang

摘要

In dynamic multiobjective optimization problems (DMOPs), quickly tracking the Pareto fronts under various changing patterns has been a key challenge. Prediction driven methods have shown strong potential in capturing these changing patterns; however, most existing approaches rely on linear models, which limits their effectiveness in addressing nonlinear changes. To address this issue, we propose a novel dynamic multiobjective evolutionary algorithm named LDFP-DMOEA that combines a linear strategy with a diffusion model. Specifically, the population is adaptively divided into two subpopulations according to the correlation of individual variations across consecutive environments. A linear prediction strategy is applied to the highly correlated subpopulation to exploit stable trends, while a diffusion model is employed for the less correlated subpopulation to capture complex nonlinear dynamics. The two subpopulations are then merged, followed by a local adjustment mechanism to enhance diversity. Experimental results show that LDFP outperforms five representative comparative algorithms on multiple benchmark problems, validating its effectiveness and superiority.