Abstract: Aiming at problems such as low evolutionary efficiency and redundant immune network which are caused by the algorithm design for Artificial Immune Recognition System(AIRS), a new artificial immune classifying method is proposed. The implementation of this method include: positive selection and network suppression are applied with Monte Carlo method to generate optimal distributing initial antibodies according to training antigens. Immune response process is imitated by using many kinds of immune mechanisms to evolve mature memory cells from initial antibodies. Classification is accomplished by majority vote of the K nearest memory cells. The classifying results for UCI data set demonstrate that, compared with AIRS, the antibodies evolutionary iterations and the network compression ratio of the new method respectively reduces 63.1% and rises 14.7% on average. In the fault diagnosis experiment for a linear stabilized voltage supply, the average classification accuracy ratio of the new method is 92.5%, which higher than the result of AIRS and the neural network.

Key words: Artificial Immune System(AIS), Monte Carlo method, positive selection, network suppression, fault diagnosis

摘要： 人工免疫识别系统的传统方法易造成抗体进化效率低、免疫网络冗余。为解决该问题，提出一种新型的人工免疫分类方法。引入阳性选择和网络抑制机理，结合蒙特卡洛方法，根据训练抗原产生优化分布的初始抗体，融合多种免疫原理模拟免疫应答过程，由初始抗体进化出成熟的记忆细胞，利用记忆细胞依据K最近邻表决方法对待分类抗原进行分类。对UCI数据集的分类结果表明，该方法与人工免疫识别系统相比，抗体进化迭代次数平均减少63.1%，网络压缩率平均提高14.7%。在某线性稳压电源的故障诊断实例中，该方法的平均分类准确率为92.5%，高于人工免疫识别系统和神经网络等分类方法。

关键词: 人工免疫系统, 蒙特卡洛方法, 阳性选择, 网络抑制, 故障诊断

CLC Number: 

• TP277