Study On The Calibration Model Transfer Method Of Near-infrared Spectrometer

In recent years,near infrared spectroscopy analysis technology has been developing rapidly,it is characterized by nondestructive testing and rapid real-time,has been applied in many fields such as industry,agriculture,medicine and so on.In actual measurement,the instrument of aging,measurement of environmental changes and other reasons,the multiple one instrument at a time to establish the calibration model and the application in other time may not apply;or,in the establishment of a multi instrument calibration model,used in another instrument when the measurement errors are larger.Therefore,in order to improve the applicability and prediction accuracy of near infrared spectroscopy(NIR)measurement models,model transfer(also called model standardization)is the key to the successful application of NIR spectroscopy.This paper focuses on the modeling of the multivariate calibration model transfer method of near infrared spectrometers.The research mainly includes the following three aspects:One,the modeling method of fast multivariate calibration model is studied.A fast partial least squares multivariate calibration model based on kernel function transformation is established--KPLS,which is used to improve the accuracy and speed of the measurement model.In this paper,the effect of the KPLS method is verified by the near-infrared experiments of human plasma in vitro,in vivo near infrared and near infrared experiments,and is compared with partial least squares(PLS)regression method.The experimental results show that:(1)human plasma in vitro experiments in the near infrared,KPLS model and PLS model the application of the prediction accuracy is close to human body;(2)in the near infrared experiment,the prediction effect of KPLS model PLS model is better;(3)in the near infrared grain experiment,KPLS model and PLS model The application of precision approach;(4)in isolated plasma near infrared experiment data,especially for large data,the calculation speed compared with the KPLS model PLS model has obvious fast.Two,study on the optimization method of sample two,transfer model,proposes a model based on multivariate curve resolution from(SIMPLISMA)a new method of singular sample and the characteristics of the sample and the combination of KPLS and SIMPLISMA-KPLS were selected,representative samples are selected to help improve the model transfer effect.In this paper,the application effect of the method is verified by the experimental data of grain near infrared,and compared with the KS-KPLS method.The experimental results show that:(1)in the 80 groups of grain samples near infrared experiment,SIMPLISMA and KS model transfer effect is close,indicating that SIMPLISMA can be used as the model in the transfer process of another new sample for researchers to choose selection algorithm;(2)to verify the combination of KPLS and 3 Sigma criteria for model calculation,at the same time in the calculation of model validation can eliminate the singular samples.Three,establish a rapid transfer model SIMPLISMA-PDS-KPLS model algorithm to improve the robustness of the near infrared spectroscopy model(or application)and near infrared detection speed,this paper verified by the data of grain near infrared experiments,and were compared with the KS-PDS-KPLS algorithm.The experimental results show that:(1)in the near infrared experiment of 80 groups of grain samples,the RMSEP of the four components after the transfer of the model is close to the modeling of the host.(2);in addition,alternative iterative complex calculation in traditional PLS algorithm is used to calculate the multiplication and addition of KPLS in the algorithm,significantly simplifies the algorithm and improve the computing speed.The SIMPLISMA-PDS-KPLS method established in this paper is of great application value for the final realization of fast and accurate on-line detection of the content and properties of substances in various fields.