基于周期的一阶隐马尔可夫模型及其学习算法

The First-Order Hidden Markov Model Based on Periodicity and Its Learning Algorithm

对于周期变动明显的数据,本文通过引入周期状态转移矩阵,提出了一种基于周期的一阶隐马尔可夫模型,分析并给出了该模型的似然计算、隐状态估计和模型训练算法。最后的数值实验表明,该模型能有效提高预测的精确度和模型的似然度,并加快模型训练的收敛速度。

By introducing a periodic state transfer matrix,a first-order HMM model based on periodicity is proposed for the data with explicit periodicity. The algorithms,including likelihood computation,hidden state estimation and model training,are also presented and analyzed. Finally,numerical simulation experiments show that the new method efficiently improves the prediction precision and the model likelihood. Moreover,it also increases the convergence rate of model training.