计算机工程与应用 ›› 2024, Vol. 60 ›› Issue (5): 76-87.DOI: 10.3778/j.issn.1002-8331.2306-0252

• 理论与研发 • 上一篇    下一篇

三阶段自适应采样和增量克里金辅助的昂贵高维优化算法

顾清华,刘思含,王倩,骆家乐,刘迪   

  1. 1.西安建筑科技大学 管理学院,西安 710000
    2.西安建筑科技大学 西安市智慧工业感知计算与决策重点实验室,西安 710000
    3.西安建筑科技大学 资源工程学院,西安 710000
  • 出版日期:2024-03-01 发布日期:2024-03-01

Expensive High-Dimensional Optimization Algorithm with Three-Stage Adaptive Sampling and Incremental Kriging Assistance

GU Qinghua, LIU Sihan, WANG Qian, LUO Jiale, LIU Di   

  1. 1.School of Management, Xi’an University of Architecture and Technology, Xi’an 710000, China
    2.Xi’an Key Laboratory for Intelligent Industrial Perception, Calculation and Decision, Xi’an University of Architecture and Technology, Xi’an 710000, China
    3.School of Resources Engineering, Xi’an University of Architecture and Technology, Xi’an 710000, China
  • Online:2024-03-01 Published:2024-03-01

摘要: 代理辅助进化算法已广泛应用于求解代价高昂的多目标优化问题,但大多数由于代理模型的局限性而仅限于解决决策变量低维的问题。为了解决高维的昂贵多目标优化问题,提出了一种基于三阶段自适应采样策略的改进增量克里金辅助的进化算法。该算法使用改进的增量克里金模型来近似每个目标函数,此模型的超参数根据预测的不确定性进行自适应更新,降低计算复杂度的同时保证模型在高维上的准确性;此外,在模型管理方面提出一种三阶段自适应采样的策略,将采样过程分为不同的优化阶段以更有针对性的选择个体,能够首先保证收敛性,提高算法的收敛速度。为了验证算法的有效性,在包含各种特征的两组测试问题DTLZ(deb-thiele-laumanns-zitzler)、MaF(many-objective function)和路径规划实际工程问题上与最新的同类型算法进行实验对比,结果表明该算法在解决决策变量高维的昂贵多目标优化问题上具有较强的竞争力。

关键词: 昂贵优化, 多目标优化, 决策变量高维, 代理辅助进化算法, 增量克里金模型, 三阶段自适应采样策略

Abstract: Surrogate-assisted evolutionary algorithms have been widely used to solve costly multi-objective optimization problems, but are often restricted to solving problems with low-dimensional decision variables due to the limitations of the surrogate model. In order to solve the expensive high-dimensional multi-objective optimization problem, an improved incremental Kriging-assisted evolutionary algorithm based on the three-stage adaptive sampling strategy is proposed in this paper. The algorithm uses an improved incremental Kriging model to approximate each objective function. This model’s hyperparameters are adaptively updated according to prediction uncertainty, reducing computational complexity while ensuring the model’s accuracy in high dimensions. In addition, the three-stage adaptive sampling strategy is proposed for model management, which divides the sampling process into different optimization stages to tailor the selection of individuals. It is able to guarantee convergence first and improve the convergence speed of the algorithm. To verify the effectiveness of the proposed algorithm, experiments are conducted on two sets of test problems containing various features DTLZ (deb-tiele-laumanns-zitzler), MaF (many-objective function) and path planning real engineering problems compared with state-of-the-art algorithms of the same type. The results show that the algorithm is highly competitive in solving expensive multi-objective optimization problems with high-dimensional decision variables.

Key words: expensive optimization, multi-objective optimization, high-dimensional decision variables, surrogate-assisted evolutionary algorithm, incremental Kriging models, three-stage adaptive sampling strategy