Research On The Modeling Of Cement Raw Meal Composition Based On Near-Infrared Spectroscopy

China is the world’s largest cement producer,and its cement output ranks first in the world for many years.The cement industry is a typical process industry.Fluctuations in the quality of raw meal during the cement production process will have a certain impact on subsequent links.Domestic cement companies mostly use manual sampling laboratory tests to obtain the composition information of cement raw meal.This method has serious time lag problems,and it is unable to timely feedback the raw meal quality changes in the production line;There are also a few companies that use neutron activation online analyzers for product composition detection,but neutron activation online analyzers have high maintenance costs and have radiation risks,which limits the application of the device.Near-infrared spectroscopy detection technology has developed rapidly in recent years,and has the advantages of rapid,non-destructive,and pollution-free detection,and has been promoted in many fields.This article takes the near-infrared spectroscopy of cement raw material and its component content as the research object.Based on the near-infrared spectroscopy detection technology and quantitative analysis method,the quantitative model of near-infrared spectroscopy of the four main oxide components of cement raw meal is established.The specific work is as follows:First of all,in a laboratory environment with relatively stable temperature and humidity,the near-infrared spectrum collection of cement raw meal powder was completed,the influence of the raw meal sample density on the raw meal sample spectrum during the spectrum acquisition process was analyzed,and the optimum sample density of the cement raw meal spectrum collection is given.Then,the principal component analysis-Mahalanobis distance method is used to remove the sample spectrum abnormally,and the SPXY method is used to divide the sample set;Tested the prediction ability of the spectral model to the four oxide content after processing by various pretreatment methods.According to the results,the first-order derivative method of Savitzky-Golay was used to process the spectrum for SiO₂,Fe₂O₃ and CaO,and the multivariate scattering correction process was used for the spectrum for Al₂O₃.Four band selection methods were applied to screen the characteristic band of cement raw meal spectrum.Among them,Interval Partial Least Squares（IPLS）,Successive Projections Algorithm（SPA）and Uninformative Variables Elimination（UVE）have good performance for different oxide components.At last,the effects of the three modeling methods of Principal Component Regression（PCR）,Partial Least Squares Regression（PLSR）,and Support Vector Machine Regression（SVMR）in the prediction of cement raw material spectral components are compared.The overall evaluation of the model performance is partial least squares regression optimal,support vector machine regression suboptimal and principal component regression worst.Combining the results of pretreatment and band selection,the prediction models of the composition of the four oxides were established.The results show that the performance of the SiO₂ PLSR model and the Fe₂O₃ PLSR model established by the spectrum after SGD-（IPLS+UVE）-SPA treatment achieves the best performance.The prediction root mean square error is 0.1359 and0.0295,and the prediction decision coefficients are 0.902 and 0.871;The performance of the Al₂O₃ PLSR model and the CaO PLSR model established by the spectrum after SPA treatment reached the best performance.The root mean square errors of the predictions were 0.0753 and0.1355,and the prediction decision coefficients were 0.849 and 0.861.