融合聚类和小生境搜索的多模态多目标优化算法

PDF下载HTML

Title:

A multimodal multi-objective optimization algorithm with clustering and niching searching

作者:

顾清华^1,2,3, 唐慧^1,2, 李学现^2,4, 江松^1,2,3

1. 西安建筑科技大学 资源工程学院, 陕西 西安 710055;
2. 西安市智慧工业感知计算与决策重点实验室, 陕西 西安 710055;
3. 西安优迈智慧矿山研究院有限公司, 陕西 西安 710055;
4. 西安建筑科技大学 管理学院, 陕西 西安 710055

Author(s):

GU Qinghua^1,2,3, TANG Hui^1,2, LI Xuexian^2,4, JIANG Song^1,2,3

1. School of Resources Engineering, Xi’an University of Architecture&Technology, Xi’an 710055, China;
2. Xi’an Key Laboratory for Intelligent Industrial Perception, Calculation and Decision, Xi’an 710055, China;
3. Xi’an Youmai Intelligent Mining Research Institute, Xi’an 710055, China;
4. School of Management, Xi’an University of Architecture&Technology, Xi’an 710055, China

关键词:

多模态多目标优化; 鸽群优化算法; 聚类策略; 小生境搜索; 非支配排序; 精英学习策略; 多样性; 地图测试应用

Keywords:

multimodal multi-objective optimization; pigeon-inspired optimization algorithm; clustering strategy; niching searching; non-dominant sorting; elite learning strategy; diversity; map-based testing application

分类号:

TP301

DOI:

10.11992/tis.202204040

摘要:

针对多模态多目标优化中种群多样性难以维持和所得等价Pareto最优解数量不足问题，提出一种融合聚类和小生境搜索的多模态多目标优化算法(multimodal multi-objective optimization algorithm with clustering and niching searching, CSSMPIO)。首先利用基于聚类的特殊拥挤距离非支配排序方法(clustering-based special crowding distance, CSCD)初始化种群；引入自适应物种形成策略生成稳定的小生境，在不同的小生境子空间并行搜索和保持等价Pareto最优解；采用特殊拥挤距离非支配排序策略实现个体选优、精英学习策略避免过早收敛。通过在14个多模态多目标函数上进行测试，并与7种新提出的多模态多目标优化算法进行对比实验以及Wilcoxon秩和检验发现，CSSMPIO的总体性能优于对比算法。最后将算法用于基于地图的测试问题，进一步证明了算法的有效性。

Abstract:

In order to address the problems associated with maintaining population diversity and the insufficient number of equivalent Pareto-optimal solutions for multimodal multi-objective optimization, this study proposes a multimodal multi-objective optimization algorithm combining clustering and niche search (CSSMPIO). In the proposed algorithm, a clustering-based special crowding distance method is designed to initialize the population. Additionally, self-organized speciation is introduced to form stable niches, facilitating parallel searching and maintaining equivalent Pareto-optimal solutions. Subsequently, a non-dominated special crowding distance is introduced to realize individual selection and an elite learning strategy, circumventing premature convergence. The algorithm has been simulated using seven other state-of-the-art algorithms on 14 multimodal multi-objective optimization problems and was tested and analyzed using the Wilcoxon rank sum test. The results reveal that the general performance of CSSMPIO is superior to that of the compared algorithms. Finally, the CSSMPIO algorithm is applied to the map-based test problem, which confirms the effectiveness of the algorithm.