系统仿真学报 ›› 2026, Vol. 38 ›› Issue (10): 2939-2963.doi: 10.16182/j.issn1004731x.joss.25-1088

• 智能优化算法与自主决策 • 上一篇    下一篇

基于残差驱动的动态多目标优化算法及其应用

张欣, 王涵, 钱鹏江, 方伟   

  1. 江南大学 人工智能与计算机学院,江苏 无锡 214000
  • 收稿日期:2025-11-06 修回日期:2025-12-04 出版日期:2026-10-31 发布日期:2026-10-20
  • 第一作者简介:张欣(1992-),女,副教授,博士,研究方向为计算智能及其应用。
  • 基金资助:
    国家自然科学基金(62106088);国家自然科学基金(62473176);江苏省自然科学基金(BK20252080)

Dynamic Multi-objective Optimization Algorithm Based on Residual-driven Two-stage Prediction and Its Application

Zhang Xin, Wang Han, Qian Pengjiang, Fang Wei   

  1. School of Artificial Intelligence and Computer Science, Jiangnan University, Wuxi 214000, China
  • Received:2025-11-06 Revised:2025-12-04 Online:2026-10-31 Published:2026-10-20

摘要:

针对动态多目标优化中预测策略难以应对前期样本少与后期精度低的困境,提出了一种基于残差分析的两阶段预测动态多目标优化算法。初始阶段,若检测到环境变化,采用适配小样本且速度快的灰度预测生成新环境个体,并通过残差分析判断是否切换为最小二乘支持向量机。通过帕累托前沿聚类选取检测对象。通过随机解、原精英解、预测解及其邻域解组成新环境下的种群,提高种群多样性并加快算法收敛。仿真结果表明:该算法在环境变化时的响应速度优异,并在动态多目标车辆路由问题中能够获得更加绿色化、短耗时、低费用的路径。

关键词: 动态多目标优化, 预测策略, 灰度模型, 最小二乘支持向量机, 动态多目标车辆路由

Abstract:

A dynamic multi-objective optimization algorithm based on residual-driven two-stage prediction (RDTP) is proposed to address the dilemma of prediction strategies in dynamic multi-objective optimization struggling with insufficient samples in the early stage and low accuracy in the later stage. In the initial stage, if an environmental change is detected, grey prediction, which is fast and suitable for small samples, is employed to generate individuals for the new environment. A residual analysis is performed to determine whether to switch to a LSSVM to improve prediction accuracy as more samples become available. The Pareto front is clustered to select detection objects. The population for the new environment is composed of random solutions, old elite solutions, predicted solutions, and solutions from their neighborhood, thereby increasing population diversity and accelerating algorithm convergence. Simulation results demonstrate that the proposed algorithm achieves an excellent response speed to environmental changes and can efficiently obtain greener, more time-saving, and lower-cost paths in dynamic multi-objective vehicle routing problems.

Key words: dynamic multi-objective optimization, prediction strategy, grey model, least squares support vector machine, dynamic multi-objective vehicle routing

中图分类号: