Application And Research Of RBF Neural Network In Intrusion Detection

With the fast development of the Internet,the problem about computer network security becomes more and more outstanding.Network security techniques include router filter,firewalls,intrusion detection,audit,counteroffensive and etc. In these techniques,router filter and firewalls are static security techniques,and others are dynamic security techniques.Static security techniques work on preventing invalid access of system,but they probably bring unknown loss when genuine network attacks occur,especially novel network attacks.Therefore,some active network security defense methods and counterattack methods need to be researched.Intrusion Detection System(IDS) is a dynamic security technique,and it has become a crucial method in network security.Presently,the false alarm rate of most IDS is high and their efficiency is low.Aimed at these drawbacks of present IDS,researches on IDS models and neural network-based intrusion detection are presented in this thesis.Firstly,this thesis introduces the traditional IDS structure and the detection method and analyzes the challenge and the trend of development that Intrusion Detection Technology faced.Discuss the application of NN in Intrusion Detection System,deep research RBFNN and analyze the merit and existed questions of the application of RBFNN in IDS.This thesis uses accelerated training algorithm based on alternative fuzzy c-means and orthogonal least squares(OLS) for that OLS widely used in training has long time training and can't determine the spreads of center nodes according to the characters of data.This algorithm reduces the number of sample that participates in OLS training and solves the problem that OLS algorithm can't determine the spreads of center nodes.Finally,this thesis carries on simulative experiment by using MATLAB7.0 tool and selecting general test data KDD99-Cup Data Set in Intrusion Detection domain. The result of this simulative experiment indicates that this accelerated training algorithm reduces the training time and the size of network and enhances the network performance of classification,thus this training algorithm enhances the Intrusion Detection Rate.