控制过程异常数据的在线检测

Online Detection of Outliers of Control Process Data

针对传统小波方法在检测异常数据方面的不足及控制系统中过程数据的特点,提出了一种适用于工业控制系统的异常过程数据在线检测方法.此方法采用基于模型的小波分析残差的检测思想,考虑异常值对模型的影响,提出了带有反馈结构的RBF网络模型,从而有效降低了异常点对RBF网络准确性的影响,提高了网络的鲁棒性;采用隐马尔可夫模型分析小波系数,避免了检测阈值的设定.实验与应用证明了该检测算法比传统小波检测算法更适合于控制过程异常数据的检测.

A new method for detecting the outliers of control process data is proposed to compensate the deficiencies of the conventional wavelet methods. A wavelet method is used to decompose the fitting error of the output and its estimate, and a RBF model with backfeed structure is developed to reduce the effect of the outliers on the precision of the RBF network, thus improving the robustness of the network. HMM is introduced into the analysis of the wavelet coefficents, which can recognize the outliers without presetting the detection threshold. Experiment and application show the validity of the proposed method.