摘要： 针对动态贝叶斯转移网络的特点，以I-ACO-B为基础，提出基于蚁群优化的分步构建转移网络的结构学习算法ACO-DBN-2S。算法将转移网络的结构学习分为时间片之间和时间片内2个步骤进行，通过改进隔代优化策略，减少无效优化次数。标准数据集下的大量实验结果证明，该算法能更有效地处理大规模数据，学习精度和速度有较大改进。

关键词: 动态贝叶斯网络, 转移网络, 结构学习, 蚁群优化

Abstract: Aiming at the characteristics of dynamic Bayesian transition networks, this paper proposes a structure learning algorithm based on Ant Colony Optimization(ACO) named ACO-DBN-2S by extending the static Bayesian networks structure leaning algorithm I-ACO-B. In ACO-DBN-2S, ants select arcs from the inter-arcs between time slices before from the intra-arcs in one slice, and the interval optimization strategy is improved by decreasing the times of optimization operation. A number of experiments under standard datasets demonstrate the algorithm can handle large data, and the precision and speed of learning are improved.

Key words: Dynamic Bayesian Networks(DBN), transition networks, structure learning, Ant Colony Optimization(ACO)

中图分类号: 

• TP301.6