基于二维细胞自动机的蚁群聚类分析及应用-analysis and application of ant colony clustering based on two-dimensional cellular automata.docxVIP

摘要随着数据挖掘研究的不断深入，群体智能越来越受到研究人员关注，作为其 重要分支的蚁群聚类算法备受学者们青睐。蚁群聚类算法是受蚂蚁群体行为启发 而设计的智能仿生算法，具有群体智能的分布式、鲁棒性、易扩展性、简单性、 广泛的适应性等特点。本文对聚类、群体智能及细胞自动机等理论进行介绍，并讨论了蚁群聚类中 的 LF（Lumer Faieta）和 BM（Basic Model）模型。在此基础上，针对 LF 的不 足对其进行改进，如参数自适应，优化蚂蚁的移动策略和行为，同时结合基于实 例库推理的思想，将蚂蚁分为普通蚂蚁和专家蚂蚁。专家蚂蚁通过查询实例库来 移动和放置数据，避免了适应度和放置概率函数的计算。并给出改进算法描述， 通过若干数据集的实验表明，该算法有效地改善了聚类性能。研究了使用将一个蚂蚁和一个数据结合为一个 agent 的人工蚂蚁休眠模型。利 用 Q 学习的思想改进了 ASM 算法中 agent 的移动策略，强化 agent 选择靠近聚类 中心移动方向的动作，加快聚类算法的收敛速度，改善了聚类质量。针对带细胞簇的 ASM 算法聚类结果簇过多的问题，将一维细胞自动机簇合并 的思想应用到带细胞簇的 ASM 算法中。把较小的簇合并到最相似的大簇中，减少 并修正了聚类簇的结果。通过分别将带细胞簇合并的 ASM 算法初步应用于孤立点 检测以及雷达辐射源的新类别识别领域，验证了其有效性。关键词：蚁群聚类细胞自动机强化学习细胞簇AbstractAlong with the deep research of data mining, swarm intelligence has been received much more attention. As an important branch if it, ant colony clustering algorithm catches the eyes of researchers. Ant colony clustering algorithm is a bionic intelligencealgorithm inspired by the collective behavior of ants，which has the characteristics ofdistributing computing, robustness, simplicity, scalability and extensive applicability.This paper introduces the theories of clustering、swarm intelligence and cellular automata. Then, two different ant colony clustering algorithms LF (Lumer Faieta) and BM (Basic Model) models are discussed. On these basis, we focus on theimprovement of LF algorithm, such as parameter adaptive、optimizing ants actionstrategy and combing the case-based reasoning method which divides the ants into normal ants and expert ants, expert ants can easily move and drop data by just inquiring the case base instead of calculating the fitness and drop probability function. Thus, a new ant colony clustering algorithm which shows preferable performance is put forward and analyzed through experiments on several data sets.We choose Ant Sleeping Model for further study, in which an artificial ant called agent represents a data object. At the same time, Q-learning algorithm is introduced into the moving strategy in ASM, thus strengthen