中文摘要：

针对核模糊c一均值算法（kernelfuzzyC—means，KFCM）随机选择初始聚类中心而不能获得全局最优且在聚类中心较近或重合时易产生一致性聚类等问题，提出一种改进算法。改进算法在原目标函数中引入中心极大化约束项来调控簇间分离度，从而避免算法出现一致性聚类结果。利用磷虾群算法对基于新目标函数的KF．CM算法进行优化，使算法不再依赖初始聚类中心，提高算法的稳定性。基于距离最大最小原则产生多组较优的聚类中心作为初始磷虾群体并在算法迭代过程中融合一种新的精英保留策略，从而确保算法收敛到全局极值；通过对个体随机扩散活动进行分段式Logistic混沌扰动，提高算法全局寻优能力。使用KDDCup99入侵检测数据进行仿真实验表明，改进算法具有更好的检测性能，解决了传统的聚类算法在入侵检测中稳定性差、检测准确率低的问题。

英文摘要：

To slove deficiency of global search ability for KFCM clustering algorithm impacted by the random selection of ini- tial cluster centers and consistent clustering occured when the clustering centers closed or overlapped, this paper proposed an improved algorithm. The objective function introduced a cluster center constraint term to regulate inter-cluster separation, thus avoided consistent clustering results. Using krill herd algorithm to optimize KFCM algorithm based on the new objective func- tion,it solved the problem of KFCM depending on initial center effectively and enhanced the instability of clustering results. U- sing a new elitist strategy in the iterative process and the max-min distance method to produce many excellent clustering centers as the initial krill populations,it could ensure that the algorithm converged to the global optimal. Using the pieeewise Logistic chaotic perturbation for individual random diffusion, it accelerated the global search ability. Experiments on data sets KDD Cup 99 show that the proposed algorithm has more efficient performance which solves poor stability and low detection accuracy of the traditional clustering algorithms in intrusion detection.