基于过程数据的动态PLS建模

许多工业过程中获得的输入输出数据在时间、空间上是高度相关的,且含有测量噪声。针对此类数据,提出采用动态PLS方法,最大化输入和输出变量矩阵之间的协方差,保留大部分有用信息,去除测量噪声,把高维数据空间降维,建立较为精确的工业过程数学模型。提出对MIMO系统进行研究,推导了可直接用于控制的动态PLS模型数学公式。对一个工业过程实例进行仿真,分别应用动态PLS回归和线性回归MLR方法,得出的结果经分析验证了此算法的有效性。

Dynamic PLS Modeling Based on Process Data

In most cases, the large amount of input and output data obtained from the industry process is not independent to each other, and even in time serials. The surveying noise is always included in them. For this situation, dynamic Partial Least Squares (PLS) was proposed. PLS is a dimensionality reduction modeling technique. It maximized the covariance between the predictor (independent) matrix and the predicted (dependent) matrix for each component of the reduced space, captured most of the data variation, and projected the system into a reduced space. Since the surveying noise was ignored in PLS, a more accurate model was obtained. Considering MIMO systems, the mathematical equation was deduced, which could be used in control systems directly. The simulation of an industry modeling example based on dynamic PLS and MLR proves the validity of dynamic PLS regression.