| Sugarcane sugar-refined is one of the important economic pillar industries in southern China.It is an important guarantee for improving the quality and working efficiency of sugar products to detect the various quality indexes of the intermediate products produced in the sugar-making process.The traditional laboratory sampling analysis method is complicated and time-consuming,which cannot meet the needs of the online detection of sugar-making process.Visible-near infrared(Vis-NIR)spectroscopy has the advantages of fast and simple analysis without reagents,and it has also been shown in some prospective basic research that Vis-NIR analysis technology is suitable for the detection of sugar intermediates.Among the research topics that need to be solved,in addition to the research and development of online NIR detectors and associated automation equipment,the establishment and optimization of the Vis-NIR spectral analysis model also urgently need to be solved.In this paper,the detections of the Brix(%)and Pol(%)indicators of clear juice and clear syrup of sugarcane sugar-refined intermediate were as the research goals.And the establishment and parameter optimization studies of related Vis-NIR spectroscopic analysis models are carried out.The main content and results are as follows:1.Method research:1)Based on partial least squares(PLS)method,a multi-partition modeling system of calibration-prediction was established.And the parameters were optimized according to multi-partition comprehensive prediction error(SEP~+)to avoid data overfitting and ensure parameter stability.In addition,the independent validation set was carried out with the samples not involved in modeling to make the results objective.2)Spectral preprocessing algorithm platform based on Norris derivative filtering(NDF)was established to overcome noise interference such as spectral baseline drift and tilt,and realize large-scaled optimization of the parameters(smooth points s,derivative order d and number of differential gaps g)of moving average smoothing and differential derivation.3)Wavelength model optimization platform was established based on equidistant combination PLS(EC-PLS)regression.The parameters of initial wavelength(I),number of wavelengths(N),and number of wavelength gaps(G)were used to achieve a wide range wavelength model selection.Based on SEP~+,the model parameters were selected to effectively extract spectral information and avoid noise interference.This method combines the advantages of continuous and discrete models,and strictly covers the classical moving-window PLS(MW-PLS)method in the algorithm.4)Wavelength step-by-step phase-out PLS(WSP-PLS)was further adopted to realize the second optimization of the wavelength model.2.Modeling and validation for the analysis of Brix and Pol in clear juice and clear syrup:The NDF method was used for the spectral pretreatment,and the parameter optimization platform of Norris-PLS model was established.the appropriate NDF parameters were determined according to SEP~+.The optimal Norris parameters(d,s,g)of the four cases(Brix and Pol of clear juice,Brix and Pol of clear syrup)were(2,5,15),(2,29,8),(2,9,7),(2,11,5),respectively.Based on the Norris derivative spectra,the EC-PLS method was used to optimize the wavelength model.Further,the WSP-PLS method was used to perform the second optimization of top 10 equidistant combination models to determine the optimal EC-WSP-PLS model.The optimal wavelength combinations corresponding to the above four models were 46,19,5,and 5,respectively.Using the samples not involved in modeling,the four optimal EC-WSP-PLS models were independently validated.The root mean square error of prediction(SEP)between the predicted values and measured values for four cases were 0.367,0.364,1.106,0.899(%),respectively.the correlation coefficient of prediction(R_P)was 0.832,0.841,0.914,0.923.the relative root-mean-square error of prediction(RSEP)was 2.2,2.5,1.8,1.6(%),which reached good prediction effect.The results showed that Vis-NIR spectroscopy combined with chemometrics can be used for rapid quantitative detection of Sugarcane sugar-refined intermediates(clear juice,clear syrup).The integrated optimization method of multi-parameter for spectral preprocessing and wavelength model has advanced and applicable value.The related algorithm platform helps to optimize the parameters of the online detection model of the sugar-making process.The established wavelength models can provide valuable reference for the design of the online NIR spectrometer. |
PDF Full Text Request