基于改进蜻蜓优化多核模糊聚类算法的异常检测  

Anomaly Detection based on Improved Dragonfly Algorithm and Multi-Core Fuzzy Clustering Algorithm

作　　者：张颖[1] 彭然 ZHANG Ying;PENG Ran(Hubei Transportation Vocational and Technical College,Wuhan 430079,China;Wuhan Institute of Technology,Wuhan 430205,China)

机构地区：[1]湖北交通职业技术学院科研处,湖北武汉430079 [2]武汉工程大学土木工程与建筑学院,湖北武汉430205

出　　处：《数学的实践与认识》2021年第19期208-219,共12页Mathematics in Practice and Theory

基　　金：湖北省交通运输厅交通运输科技项目(2020-186-3-2)。

摘　　要：对模糊C-均值聚类算法(FCM)改进及在异常检测中的应用进行研究,提出了一种基于改进蜻蜓优化多核模糊聚类算法的异常检测方案.针对FCM聚类个数事先确定,对噪声、不同形状聚类鲁棒性差以及初始聚类中心敏感的缺陷,设计最佳多核聚类策略,通过采用加权多核度量和最佳聚类评价指标,在实现自适应数据聚类个数确定的同时,证明对不同聚类问题具有更好的鲁棒性;提出密度峰值聚类中心初始化机制,通过迭代计算密度峰值函数,对聚类中心进行初始化;引入蜻蜓算法(DA),对DA迭代进化机制进行改进,并将聚类中心等效为蜻蜓个体编码,充分利用DA全局寻优优势,以改善FCM聚类性能;搭建MPI并行运算架构,将最佳多核聚类策略和蜻蜓算法局部搜索更新进行分布式计算处理,以提高数据聚类的运行效率.最后,将改进蜻蜓优化多核模糊聚类算法(IDM-FCM)应用于异常检测.仿真结果表明,对于不同形状以及孤点聚类问题,IDM-FCM具有更好的聚类效果,聚类正确率提高了约19.1%,而且,基于IDM-FCM的异常检测算法具备更高的检测能力,检测成功率提高了约5.2-39.2%,误报率降低了了约70.2-92.3%.The improvement of fuzzy C-means clustering algorithm(FCM)and its application in anomaly detection are studied,and an anomaly detection scheme based on improved Dragonfly Algorithm(DA)and multi-core FCM is proposed.Aiming at the defects of FCM,such as the number of FCM clusters determined in advance,sensitive to noise,clustering problems of different shapes and initial clustering centers,the optimal multi-core clustering strategy is designed.By using weighted multi-core measurement and the best cluster evaluation index,the number of data clusters is adaptively determined,and the better robustness to different clustering problems is proved.By calculating the density peak function iteratively,the peak clustering center initialization mechanism is proposed and the clustering center is initialized.The dragonfly algorithm(DA)is introduced with the improvement of DA iterative evolution mechanism to improve the convergence efficiency and the cluster center is equivalent to the dragonfly individual code,making full use of the advantages of the improved DA global optimization to improve the clustering performance.The MPI parallel computing architecture is built to improve the efficiency of data clustering,as the multi-core clustering strategy and local search update of DA are processed in distributed computing.Finally,the improved Dragonfly Algorithm and multi-core FCM(IDM-FCM)is applied to anomaly detection.The simulation results show that IDM-FCM has a better clustering effect for different shapes and outliers,the clustering accuracy is increased by about 19.1%,and the anomaly detection algorithm based on IDM-FCM has higher detection ability,the detection success rate is increased by about 5.2-39.2%,and the false alarm rate is reduced by about 70.2-92.3%.

关 键 词：异常检测 聚类分析 模糊C-均值 蜻蜓算法 聚类性能 检测能力 

分 类 号：TP311.13[自动化与计算机技术—计算机软件与理论] TP18[自动化与计算机技术—计算机科学与技术]