Aiming at the low true-positive rate, poor adaptability, large size and redundant detector set of existent detector generation algorithms, a self-adaptive detector generation algorithm and model based on immune-softman(ISM) characteristics is proposed. The algorithm model draws lessons from the antibodies cloning mechanism and affinity variation mechanism of the biological immune system(BIS), and integrates the niching strategy and the variation and optimization operations of the detector. Compared with traditional detector generation algorithms, the proposed algorithm can decrease effectively the redundancy of detectors, minimize the size of detector set, and maintain the diversity of detectors. At the same time, by changing continuously the matching threshold of effective detectors, the algorithm can detect quickly abnormal behaviors in the scale of the non-self space by a smaller detector set. The experiment results show that the algorithm has better adaptability and higher detection efficiency and performance.