基于异构分簇花粉算法的星系光谱特征选择  

Selection of spectral features of galaxies based on heterogeneous clustering flower pollination algorithm

作　　者：胡佳佳 熊焱[1] 孙堂旺 HU Jiajia;XIONG Yan;SUN Tangwang(School of Science,University of Science and Technology Liaoning,Anshan 114051,China)

机构地区：[1]辽宁科技大学理学院,辽宁鞍山114051

出　　处：《辽宁科技大学学报》2025年第1期65-71,共7页Journal of University of Science and Technology Liaoning

基　　金：国家自然科学基金资助项目(U1731128)。

摘　　要：针对花粉算法收敛缓慢、全局搜索能力弱等问题,结合星系光谱数据高维性及冗余特征对优化算法效率的挑战,提出改进策略,在异花授粉和自花授粉阶段分别引入异构分簇策略和排斥竞争机制,设计异构分簇花粉算法(HCFPA),以提升算法收敛速度、搜索精度及对高维数据的适应性。在CEC 2022测试函数集上的实验表明,HCFPA在大部分测试函数中表现优异。针对LAMOST DR8星系光谱数据的特征选择需求,提出基于分类精度和特征数量的指数型适应度函数,利用HCFPA高效去除冗余特征。实验结果表明,改进后的算法能够显著降低数据维度并提升分类准确率,为复杂光谱数据分析提供高效解决方案。HCFPA通过异构分簇和排斥竞争机制,快速识别关键特征,提升特征选择效率,同时降低数据维度,减少计算复杂度,为星系分类、物理参数提取及演化研究提供了更高效的解决方案。In view of the shortcomings of slow convergence and weak global search capability in traditional flower pollination algorithm,combined with the challenges posed by high-dimensional redundancy character-istics in galaxy spectral data that affect algorithmic optimization efficiency,this paper proposes an improved strategy:By introducing a heterogeneous clustering strategy at the cross-pollination stage and a repulsion com-petition mechanism during self-pollination,we have developed a heterogeneous clustering flower pollination algorithm(HCFPA).This enhanced algorithm demonstrates improved convergence speed,search accuracy,and adaptability to high-dimensional data.Experimental evaluations on the CEC 2022 benchmark test set dem-onstrate HCFPA's superior performance across most test functions.To meet the feature selection requirements of LAMOST DR8 galaxy spectral data,we propose an exponential fitness function incorporating classification accuracy and feature quantity,enabling HCFPA to effectively eliminate redundant features.Experimental re-sults indicate that the improved algorithm significantly reduces data dimensionality while enhancing classifica-tion accuracy,offering an efficient solution for complex spectral data analysis.Through its heterogeneous clus-tering and repulsion competition mechanisms,HCFPA demonstrates rapid identification of key features,im-proved feature selection efficiency,and reduced computational complexity.These advancements provide an optimized framework for galaxy classification,physical parameter extraction,and evolutionary studies.

关 键 词：花粉算法 异构分簇 特征选择 星系光谱 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]