针对现有动态贝叶斯网络结构学习方法具有低效率和低可靠性等问题,基于变量之间的基本依赖关系和依赖分析方法进行动态贝叶斯网络结构学习。建立变量之间依赖关系草图,通过条件独立行检验去除多余的边,使用碰撞识别和条件相对预测能力确定边的方向,便可得到构成动态贝叶斯网络结构的先验网和转换网。该方法在效率和可靠性方面均具有优势。
At present,the methods of learning Dynamic Bayesian Network(DBN) structure have low efficiency and reliability.Learning dynamic Bayesian network structure is done based on the basic dependency relationship between variables and dependency analysis method.A sketch of dependency relationship between variables is built.Then the redundant edges can be got rid of by the conditional independent test.And the edges are oriented through the collision identify and the relative conditional prediction capability.Therefore,the dynamic Bayesian network structure can be established.This method has high efficiency and reliability.