Study On Near Infrared Model Optimization And Transfer For Main Components Of Pulp Raw Material

Some of the main characteristics of pulping materials,such as lignin,holocellulose,moisture content,density,type and so on,will directly affect the setting of process parameters,the amount of cooking liquid,and even the quality of the final product in the subsequent production process.Therefore,rapid and accurate detection of raw materials is of great significance for optimizing process parameters and improving product quality.However,traditional wet chemical methods are time-consuming,cumbersome and cannot be used online.Near-infrared spectroscopy is a fast,efficient,low-cost "green" analysis technology,which can greatly improve the production efficiency and product quality of the pulp and paper industry.However,this technology often encounters problems with too many spectral wavelength points in the application process,leading to complex modeling process,low model prediction accuracy,and the problem that the same model cannot be shared among different instruments.In this paper,a domestic,low-cost and portable spectrometer was used to select the characteristic wavelengths of lignin and holocellulose in mixed pulp materials,and the feasibility of sharing lignin and holocellulose models among different instruments was studied.The main research contents and conclusions as follow:(1)Study on wavelength selection of lignin and holocellulose in mixed pulping materials.A low-cost portable spectrometer was used to establish a prediction model for the content of lignin and holocellulose in a variety of pulping materials.The prediction results can basically meet the production needs,but the deviations between the predicted values and the actual values of individual samples were obviously too large,that those reached 2 or more,and these models were established with full-spectrum(751 wavelength datas),this led to a high degree of model complexity,time-consuming calculations and high costs for the development of specialized portable instruments.So competitive adaptive reweighted sampling(CARS),genetic algorithm(GA)and successive projections algorithm(SPA)were used to select the characteristic wavelengths of lignin and holocellulose.Among the three methods,SPA has the least number of wavelengths selected and the best predictive performance.15 and 23 wavelengths were selected from the whole spectrum as the characteristic wavelengths of lignin and holocellulose by SPA respectively,and the models of lignin and holocellulose were established in combination with MLR,respectively.The RMSEP of the lignin and holocellulose models were 0.9129 and 1.0323,respectively,which were 9.8% and 15.6% higher than those of the full-spectrum model,respectively,and the robustness of the models were improved sharply.(2)Research on the sharing of lignin and holocellulose models among different instruments.When the predicted models of lignin and holocellulose had been used on two different instruments,the prediction deviation was obviously increased,and the prediction deviation when used for the same type of instrument(slave 1)was smaller than that of different types of instruments(slave 2).The RMSEP of lignin on slave 1 and slave 2 were 1.7418 and 9.5169,respectively,Holocellulose RMSEP were 2.1819 and 2.4251 respectively.So direct standardation(DS),spectral space transfer(SST)and canonical correlation analysis(CCA)were used to study the possibility of transfer of lignin and holocellulose model among the three instruments,and the influence of the number of standard samples on the transfer effect was discussed.The results showed that the transfer performance of DS and CCA algorithm were similar,and the more the number of standard samples,the better the transfer effect.While the SST algorithm that didn't needs as many standard samples as the other two methods can achieve better transfer performance than DS and CCA algorithm.The RMSEP of lignin on slave 1 and slave 2 were respectively 1.1262 and 1.1247 after 20 standard samples were selected by SST transfer.The RMSEP of the holocellulose was 1.0182 and 1.1979 respectively.It can be seen that only a small amount of standard sample can be used significantly.But SST algorithm needs to determine the optimal number of principal components during transfer process.Therefore,in the actual use process,the model transfer method should be selected depending on the specific situation.