Computer Engineering and Applications ›› 2023, Vol. 59 ›› Issue (4): 165-174.DOI: 10.3778/j.issn.1002-8331.2108-0256

• Pattern Recognition and Artificial Intelligence • Previous Articles     Next Articles

Hybrid Many-Objective Evolutionary Optimization Combined with  Indexs Decomposition

LI Ling, GUO Guangsong   

  1. School of Intelligent Engineering, Zhengzhou University of Aeronautics, Zhengzhou 450046, China
  • Online:2023-02-15 Published:2023-02-15



  1. 郑州航空工业管理学院 智能工程学院,郑州 450046

Abstract: Because of containing multiple different types of indicators in hybrid many-objective optimization problems, it is lack of effective evolutionary optimization method to solve this problem. This paper proposes a hybrid many-objective coevolutionary optimization method based on objectives decomposition. Firstly, the population implicit performance indicators is predicted by deep learning neural network. Then, the hybrid many-objective optimization problem is decomposed into several sub-problems based on the target correlation. The each sub-problem is solved by parallel evolutionary algorithm using multiple species. The final Pareto set of the optimized many-objective is achieved by archiving those sets of non-dominated solutions coming from the sub-populations. Finally, the Pareto optimal solution set is achieved by optimizing the aggregate function with individuals of those sets of non-dominated solutions coming from the sub-populations. The proposed method is applied to indoor layout optimization problem, the experimental results show that the proposed method is better than the contrast methods in convergence, distribution, extensibility and so on.

Key words: evolutionary computation, interactive, hybrid index, deep learning, indexs decomposition, high dimension, many-objective

摘要: 高维混合多目标优化问题因包含多个不同类型指标,目前尚缺乏有效求解该问题的进化优化方法。提出一种基于目标分组的高维混合多目标并行进化优化方法。采用深度学习神经网络预测种群隐式性能指标;基于指标相关性,将高维混合多目标优化问题分解为若干子优化问题;采用多种群并行进化算法,求解分解后的每一子优化问题,并基于各子种群的非被占优解构建外部保存集;采用聚合函数对外部保存集个体进一步优化,得到Pareto最优解集。在室内布局优化问题中验证所提方法,实验结果表明,所提方法的Pareto最优解在收敛性、分布性以及延展性等方面均优于对比方法。

关键词: 进化算法, 交互, 混合指标, 深度学习, 指标分组, 高维, 多目标