Online updating of NIR model and its industrial application via adaptive wavelength selection and local regression strategy

Near-infrared (NIR) spectroscopy has been widely used to estimate product quality or other key variables. The conventional updating strategy for an NIR model is based on new available samples. However, during a sampling interval, the model structure remains unchanged. To address this problem, in this article, a novel local regression strategy is proposed that can be adjusted according to process changes through wavelength selection and local regression approaches. The main idea of the presented algorithm is that for each query sample, a relevant calibration sample-set is selected, then the wavelength structure is updated and a local model is established. The performance of the method is demonstrated through an NIR dataset of gasoline, which was collected from a real gasoline blending and optimal control process. Compared with traditional partial least squares (PIS), locally weighted partial least squares (LW-PLS), and several other updating strategies, the proposed method is more accurate. (C) 2014 Elsevier B.V. All rights reserved.