摘要
提出一种信息论结合粒子群优化的贝叶斯网络结构学习算法,将约束最大信息熵作为最高评分函数,对网络结构进行复杂度约束,设计了粒子位置和速度向量的操作方法,解决单纯利用KL距离进行搜索的缺陷.在网络结构的搜索空间相对较大的情况下,该优化算法能在较短的时间内收敛,获得更准确的网络结构.仿真实验结果表明,该算法在时间和精度上都具有较好的效果.
A Bayesian networks learning was put forward based on information theory with particle swarm optimization algorithm. With the information entropy as the highest scoring function, the network structure complexity was constrained, and particle position and velocity vector operation designed, to solve the defects in using KL distance alone for search. In relatively large search space in the network structure, the optimization algorithm can obtain convergence in a short period of time to achieve fairly accurate network structure, and the algorithm and validation implemented through simulation experiment. The experimental results show that the algorithm has good effects in time and for precision.
出处
《厦门理工学院学报》
2014年第5期46-50,共5页
Journal of Xiamen University of Technology
关键词
贝叶斯网络
结构学习
最大信息熵
粒子群优化
Bayesian network
structure learning
maximum information entropy
particle swarm optimization
