计算机应用 ›› 2015, Vol. 35 ›› Issue (3): 685-690.DOI: 10.11772/j.issn.1001-9081.2015.03.685

• 先进计算 • 上一篇    下一篇

基于改进多目标萤火虫算法的模糊聚类

朱书伟, 周治平, 张道文   

  1. 江南大学 物联网工程学院, 江苏 无锡 214122
  • 收稿日期:2014-09-30 修回日期:2014-10-19 出版日期:2015-03-10 发布日期:2015-03-13
  • 通讯作者: 朱书伟
  • 作者简介:朱书伟(1990-),男,江苏东台人,硕士研究生,主要研究方向:数据挖掘、人工智能;周治平(1962-),男,江苏无锡人,教授,博士,主要研究方向:智能检测、自动化装置、网络安全;张道文(1989-),男,山东淄博人,硕士研究生,主要研究方向:数据挖掘、人工智能
  • 基金资助:

    国家自然科学基金资助项目(61373126);江苏省产学研联合创新资金-前瞻性联合研究项目基金资助项目(BY2013015-33)

Improved multi-objective firefly algorithm-based fuzzy clustering

ZHU Shuwei, ZHOU Zhiping, ZHANG Daowen   

  1. School of Internet of Things Engineering, Jiangnan University, Wuxi Jiangsu 214122, China
  • Received:2014-09-30 Revised:2014-10-19 Online:2015-03-10 Published:2015-03-13

摘要:

针对传统的模糊聚类算法大都针对单一目标函数的优化,而无法获得更全面、更准确的聚类结果的问题,提出一种基于改进多目标萤火虫优化算法的模糊聚类方法。首先在多目标萤火虫算法中引入一种动态调整的变异机制以获得更加均匀分布的非劣解,其中以动态减小的概率选择个体并采用类似于差分进化算法中变异算子的策略对其进行变异,通过自适应调整收缩因子以提高变异效率。然后当归档集中的最优解集充满时,从中选取一定量的解与当前种群组合进行下一次进化,使得算法具有更高的效率。最后将其运用到模糊聚类问题中,通过同时优化两个模糊聚类指标的目标函数并从最终的归档集中选取一个解确定聚类结果。采用5组数据进行实验的结果表明,相对于单目标聚类方法,所提方法对各种数据集的聚类有效性指标提高了2到8个百分点,具有更高的聚类准确性和更好的综合性能。

关键词: 模糊聚类, 多目标优化, 萤火虫算法, 变异, 差分进化

Abstract:

It has been shown that most traditional fuzzy clustering algorithms only optimize a single objective function, hence more comprehensive and accurate clustering result cannot be achieved. To solve this problem, a new fuzzy clustering technique based on improved multi-objective Firefly Algorithm (FA) was proposed. Firstly, a mutation mechanism with dynamically decreasing probability which was similar to the mutation operator in Differential Evolution (DE) algorithm was drawn into FA, in order to obtain more uniformly distributed non-dominated solutions, simultaneously the scaling factor was adaptively adjusted to enhance the efficiency of mutation. When the archive was filled, some solutions in it were selected to combine with the current population for the next evolution to improve the efficiency of the algorithm. Finally, this algorithm was applied to fuzzy clustering problem, which simultaneously optimized two objectives of fuzzy clustering index, and one solution was selected from the final archive to get the result of clustering. The experimental results on five groups of data show that the proposed algorithm raises the clustering validity index by 2 to 8 percentages than traditional single objective clustering algorithm, so it can achieve higher accuracy of clustering and obtains better comprehensive performance.

Key words: fuzzy clustering, multi-objective optimization, Firefly Algorithm (FA), mutation, Difference Evolution (DE)

中图分类号: