混合优化的贝叶斯网络结构学习  

Hybrid Optimized Algorithm for Learning Bayesian Network Structure

作　　者：许丽佳[1,2] 黄建国[1] 王厚军[1] 龙兵[1] 

机构地区：[1]电子科技大学自动化学院,成都610054 [2]四川农业大学信息与工程技术学院,雅安625014

出　　处：《计算机辅助设计与图形学学报》2009年第5期633-639,共7页Journal of Computer-Aided Design & Computer Graphics

基　　金：国防基础科研(A1420061264);国家自然科学基金(60673011);博士点基金(20070614018);电子科技大学青年科技基金

摘　　要：从大型数据库中学习网络结构一直是贝叶斯网络学习的难点之一.针对此问题提出了一种混合算法,将粒子群优化法简单且全局寻优能力强的特点,以及遗传算法良好的并行计算能力进行有效的结合,以增加学习的精度和效率.最后以经典的Asia,Cancer网络为实例,并与文中算法进行比较,验证了该算法的有效性.Learning structure from large databases is one of the difficulties of learning Bayesian Networks. To cope with this problem, a new hybrid algorithm is proposed. By integrating PSO （particle swarm optimization） and GA effectively, it owns not only simply and strong global optimization of PSO, but also favorable parallel computing capability of GA. Therefore, the learning accuracy and efficiency can be increased. Finally the proposed algorithm is compared with other algorithms in typical Bayesian networks such as Asia and Cancer, experimental results show that the proposed algorithm is effective.

关 键 词：粒子群优化法 遗传算法 贝叶斯网络 

分 类 号：TP301.6[自动化与计算机技术—计算机系统结构]