基于小波变阈值去噪递推偏最小二乘的加工质量预测方法

Process quality prediction method based on wavelet denoising recursive partial least square

为准确预测在噪声干扰下的加工质量,基于小波去噪和递推偏最小二乘方法,提出了小波变阈值去噪递推偏最小二乘方法。该方法针对小波硬软阈值去噪的不足,利用小波多尺度去噪,建立了变阈值计算公式,基于两小波域的维纳滤波,实现在偏最小二乘建模前对噪声的小波多尺度变阈值处理;同时,针对递推偏最小二乘算法中的"数据饱和"现象,基于滑动窗口的原理,通过引入折息因子控制遗忘程度,构建了多调节参数的递推偏最小二乘算法。通过该方法构建了加工质量预测模型,进行加工质量的预测,最后,结合具体实例分析,验证了该方法的有效性。

To accurately predict the process quality information interfered by noise, the method of Wavelet Diagnosis Recursive Partial Least Square （WDRPLS） based on wavelet diagnosis and recursive Partial Least Square （PLS） was proposed. Aiming at shortcomings of hard and soft threshold in the wavelet diagnosis, a variable threshold formula based on the two wavelet domains of Wiener filter was proposed by using wavelet multi-scale diagnosis so that it could diagnose by wavelet multi-scale variable threshold method before PLS modeling. To deal with data saturation problem in recurrence PLS, based on the theory of sliding window, rates factor was introduced to control forget degree, and a multi-factor recursive PLS algorithm was constructed. Process quality prediction model was set up by this method, and it could predict process quality information. Finally, an application example was used to validate the effectiveness of method.