基于FCM-GRNN聚类的入侵检测算法研究  

A Research of Intrusion Detection System Based on FCM-GRNN Clustering

作　　者：薛潇[1] 刘以安[1] 阚媛[1] 魏敏[1] 

机构地区：[1]江南大学信息工程学院,江苏无锡214122

出　　处：《计算机仿真》2010年第6期151-154,共4页Computer Simulation

摘　　要：在研究网络安全问题中,针对传统的模糊C均值聚类算法(FCM)在海量的入侵检测数据中容易陷入局部最小值,运行效率低下以及结果稳定性差的缺点,提出了一种FCM和广义回归网络(GRNN)相结合的入侵检测算法。根据GRNN的高速全局寻优特点,利用FCM将原空间的待分类样本进行聚类,利用距离FCM聚类中心最近的样本点训练GRNN模型并更新中心点,直至得到稳定的聚类中心。为解决传统的FCM在入侵检测中结果稳定性差和收敛性差,检测精度低的问题。经仿真实验结果证明,结合的方法有效的克服上述缺点,提高了数据的检测率和稳定性。To solve the problem that conventional fuzzy C -means（ FCM ） clustering algorithm in the mass invasion examination data is easy to fall into the local minimu, low in operation efficiency and poor in the result stability, the paper proposed one kind of FCM and the GRNN union invasion examination algorithm. According to the charac- teristic of the overall situation optimization and the best optimization characteristic of GRNN, this algorithm , uses FCM to carry out the clustering to the samples in original space , and uses the sample points which are most close to the FCM cluster centers to train GRNN models repeatedly and renews the central points until obtains the stable cluster centers. The purpose is to overcome the shortages of FCM, such as, poor stabilility, weak convergence and low intrusion detection precision. The experimental result proves that FCM - GRNN algorithm enhances data detection rate and stability.

关 键 词：入侵检测 神经网络 广义回归神经网络 

分 类 号：TP393.08[自动化与计算机技术—计算机应用技术]