Intrusion Detection Based On Immune Memory And Rbf Group

Modern large-scale structure of the network become increasingly complex, and it's scale grow fastly. Simply use the traditional technique based on fuzzy reasoning or expert system in intrusion detection system can not satisfy real-time and accuracy requirements.Necessary to study new intelligent detection technology, neural network and artificial immune method are the study hot. Neural network has an excellent nonlinear approximation properties, ability to study and summarized, but the neural network with limited capacity to store information, study slow,and easy to fall into local minimum. Artificial immune method has global convergence,associateive memory, self-organization,self-learning, dynamic mediation and other excellent properties.Therefore ,the combination of neural network method and artificial immune method is a good research direction, but the current study simply use immune algorithm train neural network, remain on the stage of optimise network parameters , the combination is not high enough,cause many good properties of artificial immune system, such as associative memory, self-organization, self-learning and dynamic characteristics can not be applied to the neural network. Therefore combin the neural network method with artificial immune method, play their respective advantages and come up with a better intrusion detection technology is this paper's main work and the research can be summarized as follows:1 The immune memory mechanism into the neural network method, the main idea is: First of all, through the clustering,the data will be divided into a number of clusters, then for each cluster, use a small neural network to learn, so that on the whole form a network groups, each consider a immune cell. This process is equivalent to first time immune response. The process of detection is equivalent to the secondary immune response. If there is a new data, can add a new neural network to learn the data. Small neural networks can be dynamically increased, so that solve the neural network capacity constraints. At the same time, because of the reduced network size, also increase the network learning speed. 2 Combin immune network mechanism and the on-line nuclear clustering method,proposed on-line nuclear clustering algorithm based on immune network. It can solve problems such as a number of clustering algorithms appear center drift in the cluster, the fuzzy boundary between-class and non-uniform density distribution of data between the data clusters. Because it's on-line clustering algorithm,so it's faster than the adaptive radius immune network clustering algorithm. Simulation has been done on the artificial data sets and IRIS,the result is good. Finally the algorithm as the core algorithm be used in the intrusion detection system。3 In this paper, construct a intrusion detection system based on immune memory RBF group. Given the basic idea and algorithm design, as well as the study of data preprocessing methods. This paper do a simulation test on data sets of KDDCUP99, get detection rate of 98.57 %, false positive rate of 4.56%, and achieved good effect, prove the effectiveness of the method in a certain degree.4 RBF network learning algorithm is an important factor to network performance. This paper discusses several different kinds of RBF network learning algorithm, and point out its strengths and weaknesses,and also do a simulation test. Finally select Extended Rival Penalized Competitive Learning algorithm as the paper's RBF learning algorithm.