Based On The Principal And Subordinate Self-Set And Twice Trained Artificial Immune Model

"Self-NonSelf"(SNS)recognition model is an important branch of artificial immunology models. The American immunologist Forrest provoked the negative selection algorithm based on the SNS model and successfully simulated the procedure of immunological tolerance. However, the research of Kim demonstrated the antibody set made by negative selection algorithm would produce wide zooming problems when facing large amount of detected data in the factual application network environment. The cause of zooming problem is the incompleteness of Self-Set which also named "Self-set incomplete" problem. In the traditional immunology, Self-Set always go through artificial search whereas Self set is a abnormal huge set. These all affect the cover proportion to NonSelf set of antibody by making the incompleteness of Self Set.In term of solving the problem of "Self-set incomplete" that exists in negative selection algorithm based on artificial immune, this paper designs based on the principal and subordinate structure Self-set and twice trained artificial immune model. The new modle consist of Based on the principal and subordinate structure Self-set antibody-producing algorithm (PASA), the second training algorithm (STA) and based on computed affinity PASA (CAPASA). PASA achieve the dynamic expansion of the Self-set by adding the subordinate Self-set and the antibodies generated by immune tolerance enhance the space of antigen recognition space. In STA, the generated antibodies will be re-trained in order to phase out elements of Self-set in antibodies and eliminate false-positive of antibodies. CAPASA, by calculating the affinity between antibodies, eliminate higher affinity of elements in antibodies to achieve to enhance the space of antigen recognition and to raise antibody efficiency in a limited number of antibodies. Experimental results show that this algorithm is effective and can improve the recognition performance of the detectors.