用动态Bayesian网络建立宏观经济系统模型

针对限制动态 Bayesian网络方法应用的 Markov假设和转移概率时不变假设 ,研究了如何利用部分观测信息建立宏观经济系统的 Markov模型以及如何建立转移概率具有时变特性的宏观经济系统模型。对不满足 Markov假设的演化过程 ,通过在模型中添加隐藏变量建立 Markov模型 ,并对 EM- EA算法进行扩展 ,使之用于带隐藏变量的动态 Bayesian网络的学习。对不满足时不变性的转移概率 ,应用多项式拟合方法直接从数据构造时变转移概率模型。理论分析表明了论文方法的正确性和可行性

Building macroeconomic system models with DBNs

The Markov assumption and the time-invariant assumption have limited the application of Dynamic Bayesian Networks (DBNs) to macroeconomic system modeling. The paper presents a Markov model for use with partial information and a time-variant transition probability model. For evolutionary processes that do not satisfy the Markov assumption, hidden variables are added to the evolutionary process to build the Markov model, with the EM-EA algorithm expanded for the DBNs to learn with the hidden variables. For transition probabilities that are not time-invariant, the time-variant transition probabilities are learned directly from the dataset by using a polynomial fitting algorithm. The theoretical analysis demonstrates the validity of the two methods.