部分状态可见的隐马尔可夫模型状态序列的估计方法

本文考虑了部分状态可见的隐马尔可夫模型的状态序列估计问题，在分析了现有算法无法合理估计状态路径之后，以状态转移概率、观测概率和可见状态作为先验信息，通过贝叶斯分析计算可见状态前后向状态的后验概率，并给出初始条件和递推公式，运用动态规划递推得到每个观测值对应的最可能状态以及最可能的状态路径。最后，本文给出一个系统故障识别的应用例子，验证了所设计算法的可行性。

Methods for State Sequence Estimation of Hidden Markov Model with Partially Visible States

In this paper, problems of state sequence estimation in HMM with partially visible states are studied. After analyzing the shortcoming of the existing algorithms when handling state estimation in this type of HMM, this paper calculates the posterior probabilities of the states before and after the visible state via Bayesian analysis by treating state transfer probabilities, observation probabilities and visible states as prior information, presents the initial condition and recursion formula, and obtains the most possible state corresponding to every observation symbol as well as the best estimation of the state path by using dynamic programming recursion. An application example of failure recognition is given and the proposed algorithm is verified to be feasible.