Model Optimization For Sugar Of Rice Wine Based On Near Infrared Spectroscopy

Our research group has devoted to developing the special instrument on detection components of Rice Wine. But the precision of total sugar has not reached to the national standard by developed instrument. Therefore, It has become a very important thing to improve the detection accuracy of total sugar.Shaoxing Rice Wine are taken as the research object and optimized total sugar model by dimension reduction algorithms; The best effect of total sugar model by dimension reduction algorithm are compared to the common BP model. Lastly, We made up the mixed solution according to the composition and ratio of alcohol, acid, sugar Rice Wine; analysised that alcohol, acid has been effect on total sugar. The following is the optimization ideas and conclusions:Firstly, the artile distinguished the relationship of total sugar and spectra by combination algorithm of Augmented Partial Residal Plots and runs test; Secondly, The artile reduce the dimension of the sample data by PCA, MDS, LLE, LTSA, ISOMAP, Fast ICA dimensionality reduction algorithms and built PLS model. The artile analysed each parameter in terms of impact the model, and compare the model result about each algorithm combined with PLS. Lastly, the effect of better ISOMAP, LTSA nonlinear dimensionality reduction algorithm combined with PLS model were compared to the most common nonlinear BP model, thus providing the feasibility of using dimensionality reduction algorithm to optimize the model of total sugar.According to composition distribution and proportion of alcohol and total sugar in Rice Wine, the mixed solution is made.Nextly, effects of alcohol on total sugar are analyzed. Taking alcohol concentration as variables, Their ownership of every spectral peak in mixed solution are determined and Two-dimensional infrared correlation spectroscopy is used to analyse the influence of alcohol on the spectral peak. C-H, O-H groups of alcohol and O-H, C-H of total sugar are come into being hydrogen bonds so that affect spectrum. With alcohol increased from 9.5% to 17.5%, the effect of the model RMSEP increased from 0.0108 to 0.017. It can be conclused that the result of forecast model become worse with the increase alcohol concentration.According to composition distribution and proportion of acid,total sugar in Rice Wine, the mixed solution is made and effects of alcohol on total sugar are analyzed Taking acid concentration as variables, Their ownership of every spectral peak in mixed solution are determined and Two-dimensional infrared correlation spectroscopy is used to analyse the influence of alcohol on the spectral peak. O-H,C-O groups of acid and O-H, C-O of total sugar are come into being hydrogen bonds so that affect spectrum. With alcohol increased from 3% to 7%, the effect of the model RMSEP increased from 0.048 to 0.13. It can be conclused that the result of forecast model become worse with the increase acid concentration.The reason that degree of accuracy of Rice Wine total sugar model is not high are analysed through the above methods. The relevant discussion and some ideas of optimization are proposed. It can provide some ideas about software and model to our research to develop Rice Wine special NIR instrument.