基于小波隐马尔可夫模型的控制过程异常数据检测方法  

Outlier detection for control process data based on wavelet-HMM methods

作　　者：刘芳[1] 毛志忠[1,2] 

机构地区：[1]东北大学信息科学与工程学院 [2]流程工业综合自动化教育部重点实验室,沈阳110819

出　　处：《控制与决策》2011年第8期1187-1191,1196,共6页Control and Decision

基　　金：国家863计划项目(2007AA04Z194,2007AA041401)

摘　　要：针对小波异常信号检测原理的局限性,提出了适用于过程数据的基于小波隐马尔可夫模型(W-HMM)的异常数据检测方法.首先在一定尺度下对检测信号进行分解,将频率组分不同于其他大部分信号的信号作为异常信号;然后通过计算待检测信号的小波系数与正常信号小波系数的相似概率,并利用求取隐马尔可夫模型(HMM)最优状态链的Viterbi算法对数据进行最终判断;最后通过数值验证和应用表明了所提出的检测算法的有效性和实用性.According to the limitation of the principle of outlier detection based on wavelet, this paper proposes an outlier detection method called wavelet-hidden Markov modeI（W-HMM） algorithm. In this algorithm, the signal is decomposed under some scale, and when the wavelet decompositions of the signal are different from the most other wavelet decompositions, the signal can be seen as potential outlier. Aiming to make further accurate judgement, and by calculating the similarity probability between the wavelet coefficient of this signal and that of normal signal, the final confirming is obtained by using Viterbi algorithm which is applied to HMM. Finally, experimentation and application show the effectiveness and practicality of the proposed detection method.

关 键 词：异常数据检测 改进递推小波算法 隐马尔可夫模型 过程数据 

分 类 号：TP206[自动化与计算机技术—检测技术与自动化装置]