论文部分内容阅读
隐Markov模型是一个双随机过程 ,适用于动态过程的时间序列的建模并具有强大的时序模式分类能力 ,特别适合非平稳、重复再现性不佳的信号分析 ;小波变换具有多分辨率分析的特点 ,在时频两域都具有表征信号局部特征的能力。文中将小波变换和隐Markov模型相结合 ,提出基于小波变换的HMM状态识别法 ,利用Daubechies小波进行 8尺度的小波分解 ,然后从小波分解结构中提取一维信号的低频系数作为特征向量 ,将其输入到各个状态HMM来进行训练 ,其中输出概率最大的状态即是机组运行状态 ,从而实现状态的识别 ,实验结果表明该方法很有效。
Hidden Markov model is a double stochastic process, which is suitable for the modeling of time series of dynamic process and has a strong ability of time series classification. It is especially suitable for signal analysis of non-stationary and repetitive reproducibility. The wavelet transform has multi-resolution analysis Features, in both time and frequency domain signal with the ability to characterize the local characteristics. In this paper, the wavelet transform and hidden Markov model are combined to propose a HMM state recognition method based on wavelet transform. Daubechies wavelet is used to decompose the wavelet into 8 scales. Then the low frequency coefficients of one-dimensional signal are extracted from the wavelet decomposition structure as eigenvectors. Input to each state HMM for training, in which the output probability maximum state is the unit running state, in order to achieve the state identification, the experimental results show that the method is very effective.