关键词: 人工免疫, 非选择算法, 漏洞, 检测器

Abstract: For the leaks of the available real-valued negative selection algorithm, an improved negative selection algorithm is proposed. The algorithm is based on variable-length real-valued detectors. The samples are divided into boundary samples and non-boundary samples. In the detector generation process, it identifies and records the boundary samples. While new samples is tested, whether the detectors match the boundary samples are recorded. Detailed realization and advantages of the algorithm are given. The experiments of synthetic and real data sets（Iris data set and biomedical data set） are done to test the algorithm. The results show that the algorithm has higher detection rate and need less detector numbers. It is superior to available algorithms in coverage of leaks at boundary of self and non-self.

Key words: artificial immune system, negative selection algorithm, leak, detectors