Abstract: An immune learning algorithm using real numeric code is proposed aiming at problems about anomaly detection. The constringency condition and astringency of the algorithm is studied. A method of calculating reasonable initial population of antibodies is proposed. A Reasonable range of some main parameters is presented. The experimental result indicates that the algorithm can realize optimization to distribution situation of the antibodies and clustering of data modes. High veracity of anomaly detection is obtained.

Key words: Artificial immune, Evolution and learning, Anomaly detection, Astringency

摘要： 针对异常检测问题，提出了一种基于实数编码的免疫学习算法，就算法收敛的条件、是否收敛等重要问题进行了研究；给出了算法中重要参数的取值范围。实验结果表明，提出的算法能实现对抗体分布状况的动态优化和对数据模式进行聚类，获得了较高的异常检测准确率。

关键词: 人工免疫, 进化学习, 异常检测, 收敛性