ISSN 1000-1239 CN 11-1777/TP

计算机研究与发展 ›› 2019, Vol. 56 ›› Issue (5): 1048-1059.doi: 10.7544/issn1000-1239.2019.20180099

• 人工智能 • 上一篇    下一篇

基于自适应人工鱼群FCM的异常检测算法

席亮,王勇,张凤斌   

  1. (哈尔滨理工大学计算机科学与技术学院 哈尔滨 150080) (xiliang@hrbust.edu.cn)
  • 出版日期: 2019-05-01
  • 基金资助: 
    国家自然科学基金项目(61172168);黑龙江省自然科学基金项目(F2018019);黑龙江省普通本科高等学校青年创新人才培养计划(UNPYSCT-2015048)

Anomaly Detection Algorithm Based on FCM with Adaptive Artificial Fish-Swarm

Xi Liang, Wang Yong, Zhang Fengbin   

  1. (School of Computer Science and Technology, Harbin University of Science and Technology, Harbin 150080)
  • Online: 2019-05-01

摘要: 异常检测算法在诸多领域都发挥着重要的作用.基于模糊C-均值(fuzzy C-means, FCM)的异常检测是其代表方法之一.FCM对初始值的选取很敏感,而且容易陷入局部极值.基于此的异常检测算法检测效果也不甚理想.因此,引入具有较强全局搜索能力的人工鱼群算法,对其加入自适应机制,自适应调整Visual取值范围,从而提高AFSA局部和全局寻优能力,减少算法迭代的次数.然后将其应用于FCM中,利用自适应人工鱼群算法得到的最优解进行FCM聚类分析,从而解决以上FCM存在的种种问题.最后,设计基于自适应人工鱼群FCM的异常检测算法,充分利用自适应人工鱼群的优势来提高异常检测算法的检测性能.实验表明:该算法在提高对数据的检测效率的基础上,检测性能也表现出了很好的水平,为解决异常检测模型中的检测率和虚警率相关问题提供了一种有效解决方案.

关键词: 异常检测, 模糊C-均值, 人工鱼群算法, 自适应, 全局寻优

Abstract: Anomaly detection algorithm has played a key role in many areas, and the anomaly detection based on fuzzy C-means (FCM) is one of its representative methods. Owing to the limits of FCM such as the local minimum and the sensitiveness of the selection of initial value, there is still a large room to improve the conditional FCM-based anomaly detection method. In this paper, we firstly propose an adaptive artificial fish-swarm algorithm (AAFSA), by introducing an adaptive mechanism implemented by adjusting the value range of parameter “Visual” to the artificial fish-swarm algorithm which has a strong global search ability, to improve local and global optimization abilities and reduce the times of iterations. The limits of FCM mentioned above therefore can be solved by using the optimal solution obtained from AAFSA. Then, an anomaly detection algorithm based on AAFSA-FCM is designed by making full use of advantages of AAFSA to enhance the detection performances of anomaly detection algorithm. The experimental results show that the algorithm improves the detection performance both efficiently and effectively, which provides an effective solution for solving the problems of detection rate and false alarm rate in anomaly detection models, and state-of-the-art results achieve the purpose of reducing computational costs.

Key words: anomaly detection, fuzzy C-means (FCM), artificial fish-swarm algorithm, adaptive, global optimization

中图分类号: