Near-infrared model online update method based on local weighted Lasso

Abstract

The invention provides a near-infrared model online update method based on local weighted Lasso, which belongs to the field of near-infrared on-line detection. The method is divided into four parts: historical data acquisition and preprocessing, acquisition of similarity measurement matrix, establishing of a local Lasso model and acquisition of a query sample prediction value. The method proposesthe update method based on local weighted Lasso, and a JITL method can overcome the nonlinearity of an industrial process, the generalization performance of the algorithm can be enhanced, variable selection and regression can be simultaneously carried out, the amount of calculation is reduced, the accuracy and predictability of the model can be improved, and promotion of a near-infrared on-line analysis and detection technology can be realized.

Description

technical field [0001] The invention belongs to the field of near-infrared online detection, in particular to an online update method of a near-infrared model based on a local weighted Lasso. Background technique [0002] Near-infrared spectroscopy has been widely used in many fields such as pharmaceuticals, petrochemicals, agriculture, food, and polymers due to its advantages of fast detection speed, high analysis efficiency, no or little sample pretreatment, and no damage to samples. Quantitative analysis of complex substances can be realized by establishing a model between the spectrum and the physical parameters of the measured sample. However, most of the near-infrared spectroscopy modeling is performed offline. When the offline model is put into actual production and application, its prediction performance will often be affected by factors such as process state changes, catalyst activity changes, external environmental influences, and instrument drift. Reduced, so tha...

AI Technical Summary

Problems solved by technology

However, the near-infrared spectrum has a wide wavelength range and contains hundreds of variable information. In addition to the part related to the sample properties, there are also noise redundancy that is weakly related to the sample properties or other than the sample's own information.

If all the variable information is directly used for modeling update, the performance of the model will deteriorate and the calculation cost will also increase greatly. The actual operability is not strong, which limits the promotion of near-infrared online detection technology.

Images