人工免疫分类和异常识别算法的改进

发布时间：2018-10-10 14:58

【摘要】：免疫系统是生命系统的主要系统之一,它通过从不同种类的抗体结构中构造自己与非己的非线性自适应网络,在处理复杂变化的环境中起着重要的作用。受免疫系统原理启发而发展起来的人工免疫算法具有良好的多样性、耐受性、分布式并行处理、自组织、自学习、自适应和鲁棒性等特点,能够清晰地表达学习知识,具有内容记忆功能,为学者们提供了很好的进化学习机理。它模拟的自然防御功能与异常检测系统区分正常异常的功能有着惊人的相似,并且自身具有良好的学习记忆能力和强壮的鲁棒性,因而为异常检测和分类器构造等问题的解决提供了一种新的选择。本文首先对生物免疫系统的一些基本概念、特点及机理进行了描述,介绍了人工免疫算法领域的一些经典算法及其优缺点。然后针对传统的阴性选择算法中存在的“空洞”问题,模拟抗体多样性这一特征改进算法中抗体单一性的不足,提出了一种改进的检测器大小可变的免疫异常检测算法,并通过实验分析验证了算法的有效性。针对单纯的免疫分类算法在少量训练数据下精度不高问题,引入半监督学习机制和投票决策的思想,提出了一种半监督免疫分类算法,并给出了实验分析验证了算法的有效性。
[Abstract]:Immune system is one of the main systems of life system. It plays an important role in dealing with complex changing environment by constructing self and non-self-adaptive network from different kinds of antibody structures. The artificial immune algorithm, inspired by the principle of immune system, has the characteristics of good diversity, tolerance, distributed parallel processing, self-organization, self-learning, self-adaptation and robustness, and can express learning knowledge clearly. It has the function of content memory and provides a good evolutionary learning mechanism for scholars. Its simulated natural defense function is surprisingly similar to the ability of anomaly detection system to distinguish normal anomaly, and it has good learning and memory ability and strong robustness. Therefore, it provides a new choice for solving the problems of anomaly detection and classifier construction. In this paper, some basic concepts, characteristics and mechanisms of biological immune system are described, and some classical algorithms in the field of artificial immune algorithm are introduced, as well as their advantages and disadvantages. Then an improved immune anomaly detection algorithm with variable detector size is proposed to solve the "void" problem in traditional negative selection algorithm and to simulate the lack of single antibody in the improved antibody diversity algorithm. The validity of the algorithm is verified by experimental analysis. Aiming at the problem that the accuracy of the immune classification algorithm is not high under a small amount of training data, a semi-supervised immune classification algorithm is proposed by introducing the idea of semi-supervised learning mechanism and voting decision, and the validity of the algorithm is verified by experimental analysis.
【学位授予单位】：福建师范大学
【学位级别】：硕士
【学位授予年份】：2011
【分类号】：R392.1

【参考文献】