摘要
提出了一种新的隐马尔可夫模型(HMM)拓展模型自回归隐半马尔可夫过程(Auto-Regressive Hidden Semi-Markov Model,AR-HSMM),并给出了模型参数的推导和相应的"前向-后向"算法.与传统的HMM相比,AR-HSMM有以下两个优点:①把传统HMM所假设的隐藏状态分布改进为显式高斯分布;②改进了传统HMM假设各观测变量相互独立的问题,通过在各观测变量之间建立联系,从而使之更加符合实际情况.在液压泵故障诊断中的应用实例表明,AR-HSMM在故障诊断中是非常有效的.
This paper presented a new model AR-HSMM （Auto-Regressive Hidden Semi-Markov Model） that relaxes some limits of traditional HMM and developed the parameters re-estimation formula and modified ＂Forward-Backward＂ algorithm. Compared with the traditional HMM, the AR-HSMM has two significant advantages：① It modifies the unrealistic exponential distribution for hidden states by using explicit Gaussian distribution; ② Instead of the independence assumption between observations, AR-HSMM employs auto-regression to describe the correlations between observations. The proposed model was validated by a real bump diagnosis example. Compared with the traditional HMM, the promising results from AR-HSMM show that the proposed method is effective.
出处
《上海交通大学学报》
EI
CAS
CSCD
北大核心
2008年第3期471-474,479,共5页
Journal of Shanghai Jiaotong University
基金
上海浦江计划资助项目(05PJ14067)
