摘要
针对随机搜索机制学习算法参数设置较多的不足,提出了一种基于禁忌搜索的贝叶斯网结构学习算法.此算法首先利用加边、减边、逆向边3个算子产生当前解的邻域,然后结合禁忌表和蔑视准则以引导和限制搜索过程,2个步骤迭代进行,直至达到全局最优解或近似最优解.仿真实验表明,此算法不仅具有结构简单、参数少、易于实现的特点,而且求解质量也能得到保证.
To solve the drawbacks of the random searching based algorithms for learning Bayesian networks, we introduced the Tabu search into Bayesian network structure learning problems, proposed a Tabu-search-based Bayesian network structure learning algorithm (TBN). First, the new algorithm generates the neighborhood solutions by add, subtract and reverse arc operators. And then, the Tabu list and aspiration criteria guide the search procedure corporately. After the iteration of the two steps above, the algorithm will finally obtain optimal or near optimal solutions. The experiment results on the benchmark data sets show that TBN has a simpler structure and faster speed.
出处
《北京工业大学学报》
EI
CAS
CSCD
北大核心
2011年第8期1274-1280,共7页
Journal of Beijing University of Technology
基金
北京市自然科学基金资助项目(4102010)
关键词
贝叶斯网
禁忌搜索
结构学习
Bayesian networks
tabu search
structure learning