计算机工程与应用 ›› 2021, Vol. 57 ›› Issue (23): 71-80.DOI: 10.3778/j.issn.1002-8331.2012-0010

• 理论与研发 • 上一篇    下一篇

异构集成代理辅助多目标粒子群优化算法

陈万芬,王宇嘉,林炜星   

  1. 上海工程技术大学 电子电气工程学院,上海 201620
  • 出版日期:2021-12-01 发布日期:2021-12-02

Heterogeneous Ensemble Surrogate Assisted Multi-objective Particle Swarm Optimization Algorithm

CHEN Wanfen, WANG Yujia, LIN Weixing   

  1. School of Electronic and Electrical Engineering, Shanghai University of Engineering Science, Shanghai 201620, China
  • Online:2021-12-01 Published:2021-12-02

摘要:

针对代理辅助进化算法在减少昂贵适应度评估时难以通过少量样本点构造高质量代理模型的问题,提出异构集成代理辅助多目标粒子群优化算法。该方法通过使用加权平均法将Kriging模型和径向基函数网络模型组合成高精度的异构集成模型,达到增强算法处理不确定性信息能力的目的。基于集成学习的两种代理模型分别应用于全局搜索和局部搜索,在多目标粒子群优化算法框架基础上,新提出的方法为每个目标函数自适应地构造了异构集成模型,利用其模型的非支配解来指导粒子群的更新,得出目标函数的最优解集。实验结果表明,所提方法提高了代理模型的搜索能力,减少了评估次数,并且随着搜索维度的增加,其计算复杂性也具有更好的可扩展性。

关键词: 多目标优化, 粒子群优化算法, Kriging模型, 径向基函数网络模型, 异构集成

Abstract:

Aiming at the problem that the surrogate-assisted evolutionary algorithm is difficult to construct a high-quality surrogate model from a small number of sample points when reducing expensive fitness evaluation, a heterogeneous ensemble surrogate-assisted multi-objective particle swarm optimization algorithm is proposed. This method uses the weighted average method to combine the Kriging model and the radial basis function network model into a high-precision heterogeneous integrated model to achieve the purpose of enhancing the algorithm’s ability to process uncertain information. Two surrogate models based on ensemble learning are applied to global search and local search respectively. Based on the framework of multi-objective particle swarm optimization algorithm, the newly proposed method adaptively constructs a heterogeneous ensemble model for each objective function, and then uses the non-dominated solution of the model to guide the update of the particle swarm and obtains the optimal solution set of the objective function. Experimental results show that the proposed method improves the search capability of the surrogate model, reduces the number of evaluations, and as the search dimension increases, its computational complexity also has better scalability.

Key words: multi-objective optimization, particle swarm optimization algorithm, Kriging model, radial basis function network model, heterogeneous ensemble