基于改进细菌算法的贝叶斯网络结构学习

发布时间：2018-05-02 18:36

本文选题：贝叶斯网络 + 细菌觅食优化算法　；参考：《浙江大学》2017年硕士论文

【摘要】：贝叶斯网络(Bayesian Networks,BNs)是为了解决不确定性和不完整性问题而提出的一种基于概率推理的图形化网络。随着人工智能和数据挖掘领域的兴起,贝叶斯网络也越来越成为研究的热点。对于贝叶斯网络结构学习这一NP难问题的研究更是成为了研究的重点。本文首先阐述了贝叶斯网络的起源与发展,贝叶斯网络目前的应用。结合简单的贝叶斯网络结构,介绍了贝叶斯网络的基础知识以及贝叶斯分类器的知识,并介绍了细菌算法的理论依据和研究现状。其次,根据贝叶斯网络的特点,对细菌算法做了相应的改进,在传统细菌算法的基础上,针对贝叶斯网络的特点,提出基于改进细菌觅食优化算法的贝叶斯网络结构学习策略,将传统细菌算法中的趋化算子、繁殖算子和迁移算子进行改进。将自适应理论应用于细菌游动步长的计算和繁殖个体的选择中;在迁移算子的迁移概率计算中,引入遗传算法中的轮盘赌方法;在互信息理论的基础上,给出一种新的网络结构随机进化方法,代替传统细菌算法中的随机迁移,并将改进后的细菌算法应用于贝叶斯网络结构学习中。最后,通过实验仿真并将本文算法学习得到的网络结构与贪婪算法、GTT算法和K2算法学习得到的网络结构进行比较,实验证明本文算法在贝叶斯网络结构学习中能得到较好效果。
[Abstract]:Bayesian Networks (BNs) is a graphical network based on probabilistic reasoning to solve the problem of uncertainty and incompleteness. With the rise of artificial intelligence and data mining, Bayesian network has become a hot research topic. The study of Bayesian network structure learning, a NP-hard problem, has become the focus of the research. In this paper, the origin and development of Bayesian network and the current application of Bayesian network are described. Combined with the simple Bayesian network structure, the basic knowledge of Bayesian network and the knowledge of Bayesian classifier are introduced, and the theoretical basis and research status of bacterial algorithm are introduced. Secondly, according to the characteristics of Bayesian network, the corresponding improvement of bacterial algorithm is made. According to the characteristics of Bayesian network, the learning strategy of Bayesian network structure based on improved bacterial foraging optimization algorithm is put forward. The chemotaxis operator, propagation operator and migration operator in the traditional bacterial algorithm are improved. The adaptive theory is applied to the calculation of bacterial walking step size and the selection of propagating individuals; in the calculation of migration probability of migration operator, the roulette method of genetic algorithm is introduced; on the basis of mutual information theory, the method of roulette is introduced. A new random evolution method of network structure is proposed to replace the random migration in the traditional bacterial algorithm, and the improved bacterial algorithm is applied to Bayesian network structure learning. Finally, the network structure obtained by this algorithm is compared with that of greedy algorithm, GTT algorithm and K2 algorithm. The experimental results show that the proposed algorithm is effective in Bayesian network structure learning.
【学位授予单位】：浙江大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP18

【相似文献】