| Near infrared spectroscopy(NIRS),as a process analysis technology,is widely used in the quality analysis of drugs,food and agricultural products.The analysis process is as follows:spectra were collected by near infrared spectroscopy instrument;concentration or property parameters of specific components of samples was determined with standard reference method;before the establishment of NIRS models,different preprocessing methods and variables selection method were discussed.Then the NIRS model was used to predict the concentration or property parameters of the unknown samples,and the predictive capability and effectiveness of the NIRS model were evaluated.From the above process,it can be found that the stability of the original spectrum and basic data,and the variables selection methods would affect the predictive capability of the final NIRS model.The detection of plasma protein content is an important item in the daily plasma collection and production process.Previous studies have applied NIRS to detect the protein content.However,there is no innovative method of NIRS modeling.This study aims to innovate the NIRS method for detecting the plasma protein content,and propose a strategy.The strategy includes "Data mean" and "Ratio of absorbance to concentration".Three factors of NIRS modeling were discussed.The specific research contents were as follows:(1)Improve the stability of basic data of near infrared spectroscopy analysis technology based on "Data mean" method"Data mean" method was proposed based on the principle of "statistics",when averaging the values of enough times detecting,there was no systematic error in the actual measurement,and the average value was close to the truth value.In our research,the number of parallel sampling was changed,and the stable mean value was obtained by multiple parallel sampling.The minimum times of parallel collection were obtained through the cumulative average analysis,that is,by analyzed the Matlab simulated data and the actual plasma protein content,the minimum times of parallel measurement of basic data was 39 times.Compared with the NIRS model established with the conventional basic data,the root mean squares error of prediction was increased by 23.28%(2)Improve the stability of modeling spectrum of near infrared spectroscopy analysis technology based on "Data mean" methodData mean" method was used to analysis the spectra,which was similar to the processing of basic data.That is to say,changing the conventional method of averaging three near infrared parallel spectra.By analyzing the cumulative average absorbance at different wavenumber,the minimum number of near infrared parallel spectra was determined.The relative stable near infrared spectra can be obtained by averaging 11 spectra after our research.The NIRS model,which was established by data mean method,was better than that of the model of average three parallel spectra in predictive capability.(3)Improve the predictive capability of near infrared modeling based on "Ratio of absorbance to concentration" methodRatio of absorbance to concentration" method was to analyze the change rule of absorbance with concentration under different wavenumbers of sample spectra,the larger Vmean value was used to select variables for NIRS modeling.Except the quantitative analysis of plasma protein content,in order to test the generality of new method,the public data network(corn and gasoline near infrared spectra)were also analyzed.Compared with the conventional variables selection method,ratio of absorbance to concentration method could improve the predictive capability of NIRS model.The innovations of this paper include:(1)"Data mean" method improves the stability of basic data and modeling spectrum of NIRS technology,and improves the predictive capability of NIRS model.(2)Compared with the conventional variables selection method,"Ratio of absorbance to concentration" method not only improves the predictive capability of NIRS model,but also has a wide range of versatility. |
PDF Full Text Request