摘要
K均值聚类算法对初始值的选取依赖性极大,易陷入局部极值。为此,结合模拟退火算法和K均值聚类思想,提出一种新的入侵检测方案。算法利用模拟退火算法对聚类分析中的聚类准则进行优化,以获得全局最优解,并进一步开拓模拟退火算法的并行性以加快算法收敛速度。在KDD CUP 1999上进行了仿真测试,实验结果表明该方案优于基于K均值聚类的入侵检测算法,有较低的误检率与虚警率。
Intrusion detection algorithms based on K-mean clustering have sensitive dependence on initial value and are easy to fall into local extremum. To solve this issue,a new intrusion detection scheme was presented by combing Simulated Annealing and K-mean clustering. The proposed algorithm uses SA to optimize the clustering pattern in the clustering analysis. It can achieve global optimization and better accuracy of the intrusion detection system. Moreover, parallelism of SA greatly quickened the convergence rate. Experiments were completed on KDD Cup 1999, and the results show that presented scheme has lower time consume, false positive rate, and false negative rate compared with intrusion detection systems based on K-mean clustering.
出处
《计算机科学》
CSCD
北大核心
2010年第6期122-124,共3页
Computer Science
基金
国家"973"项目子课题(2007CB310702)
湖南省自然科学基金项目(09JJ3124)
广东省自然科学基金项目(7007730)
广东省科技计划项目(0711020400157)
东莞市科技攻关项目(2006D1046
2007108101021)资助
关键词
入侵检测
模拟退火
K均值聚类
全局优化
Intrusion detection,Simulated annealing,K-mean clustering,Global optimization,Parallelism