The Research And Improvement Of Real-valued Negative Selection Algorithm With Vaeible-sized Detectors

With the prevalence of computer network, it has brought us conveniences, and atthe same time produces a lot of problems. For example, computer security is oneoutstanding problem. Network Intrusion Detection is an essential component ofinformation security infrastructure. Network Intrusion Detection System is mainlyaimed to distinguish between normal behaviors and abnormal behaviors. BiologicalImmune System (BIS) has a good robustness, diversity and adaptability, as well asdynamic, self-learning and autonomy characteristics. This provides a reference to builda reliable intrusion detection system. Based on investigation and analysis of variousalgorithms and models of artificial immunity, this paper introduces and improves theReal-Valued Negative Selection Algorithm with Variable-Sized Detectors algorithm.The original Real-Valued Negative Selection Algorithm with Variable-SizedDetectors has a relatively low correct detection rate and high false positive rate in highdimensionaldata set. To solve this problem, we improve the algorithm of detectorgeneration, through the manner of generating detectors'radius by multiplying a ratiocoefficient. After the improvement, however, the number of detectors is too large, andthis will affect the efficiency of detection. So we use the Particle Swarm Optimization(PSO) algorithm to optimize the distribution of detectors. The PSO algorithm letsdetectors distribute to the best position of the non-self space where the density ofanomaly samples is large. As a result, we only generate a small number of detectorswhile get the best detection efficiency. In the optimization process, we improve themethods of dealing with problems of collisions between particles which appear in thePSO algorithm. At the same time, we propose a new algorithm to solve the issues ofwhich detectors cover self-training samples from the bodies after the optimization. Also,we use the artificially generated low-dimensional data set and high-dimensionalKDD1999 data set to do simulation experiments. The result shows that the improvedalgorithm for these two data is very effective. It has remarkable advantages over theoriginal Real-Valued Negative Selection Algorithm with Variable-Sized Detectorsalgorithm in both correct detection rate and false alarm rate, while reducing the numberof detectors.