Research On Intrusion Detection System Based On Neural Network Antibody Group

With the growing complexity of network size, the rapid development of computer technology brought to the network security challenges:The complexity of the attacks was rising, while the level of knowledge required for the attacker was a downward trend. In response to a strong threat of cyber-attacks, there has been based on artificial intelligence, soft computing intrusion detection system thinking, in which artificial neural networks, artificial immune method for intrusion detection is a hot research. Artificial neural network optimization with high-speed, self-learning ability is a good approximator of the nonlinear relationship; the artificial immune system mimic biological immune system and intrusion detection systems; in essence, the target has its similarities:Both are against foreign invasion, to maintain the normal operation of the system itself. Therefore, artificial neural network and artificial immune system combined together for intrusion detection will help to improve the degree of imitation of living organisms to improve the performance of intrusion detection system.The current study, using immune algorithm to train a single neural network, although the use of RBF neural network to avoid falling into "local minimum" problem, but the current algorithm to mimic biological immune response in the process, Antibody did not fully reflect the stage and clear stage of antigen-specific antibodies against antigens. Therefore, many types of network attacks, a large number of cases, the trained RBF neural network difficult to cope with a single intrusion detection efficiency and accuracy requirements. How to better mimic biological immune system, artificial immune and neural networks will combine to further enhance the performance of intrusion detection system, as the main contents of this paper. The main research works are as follows:1. Mimic biological process of the initial immune response, immune algorithm for training RBF neural network is proposed for the antigenic properties of the antibody generation algorithm, a training data as an antigen, an RBF network as an antibody. In the training phase, according to a number of different types of training data to generate a number of different antibodies RBF network, and the immune clonal selection algorithm, introduce the idea of variation and vaccine RBF network training is used to determine the size of each RBF network And the center of hidden layer nodes, making the antibody with the structure and size of the antigen can adaptive changes.2. Mimic biological process of the secondary immune response, an antibody features oriented Antigen clearance algorithm is proposed. In the intrusion detection system detection phase, not all of the training RBF networks are obtained in testing, but the selection of data to be detected with the corresponding antibody in detection of RBF network, which makes the detection of increased efficiency and accuracy.3. Construct a neural network-based intrusion detection system antibody group, KDDCUP99data sets using a simulation experiment, and with the use of a single antibody immune algorithm, the performance of traditional antibody groups were compared, Experiments show that the algorithm has higher accuracy and adaptability.