Research On Multi-Indicator Quantitative Analysis Of Sugarcane Clear Syrup Based On BP Neural Network

Brix and Pol are two key indicators of clear syrup,an intermediate product of sugar cane sugar making process,so rapid detection of these two indexes is very important for the production of high-quality sugar products.However,the existing detection methods are highly specialized,complex and time-consuming,and cannot meet the demand for rapid detection of index content in the production process.Therefore,it is of great significance to develop a method for quickly detecting the index content of clear syrup for the sugar production process.The research method of artificial neural network combined with near-infrared spectroscopy can carry out rapid quantitative analysis of the substances to be measured,and has been widely used in many fields,but few scholars use it in the research on the related indicators of sugarcane clear syrup.To this end,the main work of this thesis is as follows:First,this thesis uses(Standard Normal Variate)SNV spectral preprocessing method to eliminate the noise in the spectrum of clear syrup.Considering the limitations of using full-spectrum wavelength data modeling,the characteristic wavelengths of the full-spectrum bands are screened by two wavelength screening methods(Moving Window Partial Least Squares Method)MW-PLS and(Equidistant Combination Partial Least Squares Method)EC-PLS and the characteristic wavelengths are optimized according to the PLS modeling effect.After comparative analysis,the SNV preprocessing method adopted in this thesis has a significant optimization effect on the correction of the spectral data of clear syrup.The selected characteristic wavelength model can replace the full spectrum wavelength to obtain lower model complexity and better prediction accuracy.Secondly,unlike previous methods of constructing neural network models,this thesis proposes(Backpropagation Artificial Neural Network)BPANN combined with the modeling method of characteristic wavelength,this method takes into account the representativeness of characteristic wavelengths and the learning ability of BP neural network.The characteristic wavelength is used as the input variable,and the physical and chemical values of the two indicators are used as the target variables to construct the BP neural network model,and the relevant parameters are selected for model training.After comparing the modeling results,it was shown that the BP neural network model constructed by the EC-PLS characteristic wavelength showed good predictive ability in the prediction of the multi-index content of clear syrup.Finally,all the models constructed in this thesis are independently tested using the test set samples that did not participate in the modeling to obtain more objective experimental results.The test results show that,compared with the PLS algorithm commonly used in sugarcane sugar research,the model constructed by the BP neural network algorithm in this thesis shows better prediction accuracy and robustness in the prediction of sugarcane clear syrup Brix and Pol index content sex.To sum up,this thesis applies the research method of BP neural network combined with near-infrared spectroscopy to the detection of the Brix and Pol index content of clear syrup,and the feasibility of the method is verified by experiments.Compared with traditional detection methods,the quantitative analysis model constructed in this thesis has higher detection efficiency and is more suitable for the prediction of sugarcane clear syrup Brix and Pol index content.In addition,the design of the characteristic wavelength spectrum matrix and related algorithm parameters obtained in this thesis also has certain reference significance for the development of special near-infrared instruments.