Studying A Multi-Agent Artificial Immune Model Based On Danger Theory

Most of the current artificial immune systems are built on the traditional self-nonself (SNS) discrimination model for anomaly detection. Due to its inherent limitation, SNS model is not able to solve the problems listed following when put into practice, how to stipulate the self-set without any nonself in its definition and how to make the self-set gear to the dynamic environment. Therefore, in the real world, SNS model is likely to bring about scaling problems, as well as result high false detection rate. To offset these shortages, immunologists have developed a new theory-danger model recently. It suggests that the immune responses are always triggered by danger signals rather than nonself, thus, what we really should focus on is danger but foreign entities.Standing on the shoulders of the SNS models and the existing danger models, this paper investigate a novel multi-agents immune system inspired by danger theory for anomaly detection. This new model defines the danger zones through the use of the changes of the parameters derived by the system, and then the way of an immune response to the special situation is determined by the comprehensive evaluation of the degree of danger signals sent from multiple agents lied in the network. In the process of dealing with the danger zones, immune cells are incited by the combining of the danger signals and the co-simulation signals. Consequently, T killer cells and B cells are activated by T helper cells to eliminate the danger bodies and mitigate the infection. The major deference from the former models is that the novel model is trying to categorize the danger zones logically instead of the simply unified definition, and takes multiple immune responses to control the danger zones. In this paper, we use one kind of data sets to implement the experiment, and the experimental results show high anomaly detection efficiency.