基于多重小波变换的信号去噪及其在软测量中的应用

化工生产过程中采集到的数据信号通常具有随机性和非平稳性，附加了各种噪声，以至于影响数据建模的拟合效果和泛化性能。本文基于小波分析的特点，提出了一种对信号数据进行多重小波变换阈值去噪的方法。该方法可去除大部分高频随机噪声，提取真实信号，进而提高数据的置信度。将该方法与小波神经网络相结合并应用于丙烯腈聚合反应过程质量指标软测量模型中。仿真结果表明，该方法能有效恢复数据的真实性，提高数据建模的拟合精度与泛化性能。

Signal de-noising based on multiple wavelet transform and its application in soft measurement

The data acquired in chemical production process usually feature non-stationary and random nature and contain various noises inevitably, which affects the fitting and generalization capability in data modeling. Based on the characteristics of wavelet analysis, this paper proposes a method that combines multiple wavelet transform with a new threshold function. The method can remove most random noise, extract true signal and improve the confidence level of the data. The method is combined with wavelet network and was applied to the soft measurement modeling for the performance specification of polyacrylonitrile production process. Simulation results indicate that the method can effectively recover the factuality of the data and improve the fitting and generalization capability in soft measurement data modeling.