局部主成分分析及其在软测量中的应用

Local Principle Component Analysis and Its Application in Soft Sensing

在软测量建模中,局部学习策略是解决过程时变特性及非线性的有效途径。为提高模型预报精度,提出了一种基于局部主成分分析的在线软测量建模方法。该方法采用主成分建立局部模型,同时考虑变量间的相关性以及映射关系构建局部模型的选择准则。此外,为了提高在线计算效率、降低对存储设备的要求,提出了一种基于预报残差方差和均值的冗余模型判别方法。在某连续搅拌反应器上的仿真结果验证了该方法的有效性。

In soft sensing modeling, local learning strategy is an effective way to solve the time varying and nonlinear characteristics. In order to improve the accuracy of model prediction, the online soft sensing modeling method based on local principle component analysis is proposed. With this method, the local model is buih by adopting principle component, and both the correlation relationship and mapping relationship between process variables are taken into consideration to provide appropriate selection criteria for building local model. In addition, to improve online compulational efficiency and reduce the demands for storage devices, the discriminant method of redundant models based on the residual variance and mean value of prediction is also proposed. The simulation result on certain continuous stirred-tank reactor （ CSTR ） verifies the effectiveness of this method.