An immune optimization based real-valued negative selection algorithm

摘要

Negative selection algorithms are important for artificial immune systems to produce detectors. But there are problems such as high time complexity, large number of detectors, a lot of redundant coverage between detectors in traditional negative selection algorithms, resulting in low efficiency for detectors’ generation and limitations in the application of immune algorithms. Based on the distribution of self set in morphological space, the algorithm proposed in this paper introduces the immune optimization mechanism, and produces candidate detectors hierarchically from far to near, with selves as the center. First, the self set is regarded as the evolution population. After immune optimization operations, detectors of the first level are generated which locate far away from the self space and cover larger non-self space, achieving that fewer detectors cover as much non-self space as possible. Then, repeat the process to obtain the second level detectors which locate close to detectors of the first level and near the self space and cover smaller non-self space, reducing detection loopholes. By analogy, qualified detector set will be obtained finally. In detectors’ generation process, the random production range of detectors is limited, and the self-reaction rate between candidate detectors is smaller, which effectively reduces the number of mature detectors and redundant coverage. Theoretical analysis demonstrates that the time complexity is linear with the size of self set, which greatly reduces the influence of growth of self scales over the time complexity. Experimental results show that IO-RNSA has better time efficiency and generation quality than classical negative selection algorithms, and improves detection rate and decreases false alarm rate.