基于改进粒子群优化算法的贝叶斯网络结构学习  

Bayesian Network Structure Learning Based on Improved Particle Swarm Optimization

作　　者：高晓光[1] 邸若海[1] 郭志高[1] 

机构地区：[1]西北工业大学电子信息学院,陕西西安710129

出　　处：《西北工业大学学报》2014年第5期749-755,共7页Journal of Northwestern Polytechnical University

基　　金：全国高校博士点基金(20116102110026)资助

摘　　要：贝叶斯网络结构学习是数据挖掘和知识发现领域的重要研究技术之一,在网络结构的搜索空间较大的情况下,传统的二值粒子群优化算法往往存在收敛速度慢,容易陷入局部最优,学习精度较差的缺陷。在传统二值粒子群优化算法基础上,利用互信息限制粒子群算法的初始化,缩小算法的搜索空间,同时构建新的进化模型代替原有的进化公式,使得改进后的算法具有更强的寻优能力。采用ASIA网络作为仿真模型,并与原有算法比较,结果表明,改进算法能够在较少的迭代次数下找到较优的解,并且基本没有增加算法的复杂度。Bayesian network structure learning is one of the important research techniques in the domain of data mining and knowledge discovery, when the search space of the network structure is bigger, traditional binary parti-cle algorithms often have some defects such as low convergent speed, falling easily into local optimum and low pre-cision. We improve the classic binary particle swarm optimization algorithm in two respects： particle initialization and update process;the improved algorithm has stronger optimization ability. We compare the proposed algorithm with the original algorithm using the ASIA network. The results and their analysis show preliminarily that the pro-posed algorithm is able to find the better solution with less number of iterations, without increasing the complexity basically.

关 键 词：贝叶斯网络 数据挖掘 粒子群优化 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]