Rapid Detection Of Thermal Chemical Engineering Characteristics Of Straw Based On Reflection And Transmission Spectroscopy

The thermal chemical engineering characteristic is an important characteristic of energy utilization of straw biomass. In order to explore the method for rapid detection of thermal chemical engineering characteristics of straw biomass, 188 different varieties of straw samples from different regions have been collected and prepared, including the straw of rice, wheat, canol and corn. The element analysis indexes, gross/net calorific value and industrial analysis indexes of sample have been measured according to the national standard. The near infrared diffuse reflectance spectroscopy, visible/near infrared reflectance spectroscopy and visible/near infrared transmittance spectroscopy information has been respectively obtained by 3 ways of using the Fu Liye spectrometer, reflectance and transmittance hyperspectral imaging systems with the application of diffuse reflectance and transmittance spectroscopy technology. Using the partial least squares(PLS) algorithm to respectively establish the quantitative analysis models of the thermal chemical engineering characteristics of straw biomass under the 3 ways. Comparing the models established in different pretreatment methods and variable selection algorithms based on the spectrum obtained by 3 kinds of spectral collection methods, and determine the optimal detection model of each index.The main results are as follows:(1) Comparing the calibration effect of models based on near infrared spectroscopy, reflectance and transmittance hyperspectral imaging technology, the models which detect element analysis indexes(C, H, N, S, O) based on hyperspectral reflectance technique is the optimal models.The optimal pretreatment methods of spectral are: C element is without pretreatment(None); N element is algorithm of remove the trend of transformation(Detrend); H element is algorithm of multiple scattering correction combined with standard processing(MSC+Autoscale); S element is algorithm of first-order derivative process combined with standard processing(FD+Autoscale); O element is algorithm of remove the trend of transformation combined with mean Centre(Detrend+ Mean Center). Determine the competitive adaptive reweighted sampling(CARS) for optimal variable selection algorithm. The quantitative analysis models of each element have been optimized, then the number of variables in modeling decreases significantly and the stability and predictive performance of models all have improved. The models of N and O element are the optimal models. The model of N element has been built with 24 variables, the correlation coefficient of validation set(Rp) is 0.923, the mean square root error of prediction(RMSEP) is 0.196%, the relative analysis error(RPD) is 3.11; The model of O element has been built with only 10 variables, the correlation coefficient of validation set(Rp) is 0.876, the mean square root error of prediction(RMSEP) is 1.015%, the relative analysis error(RPD) is 2.32. The models of N and O element could be used in practical applications, while the prediction effect of CARS-PLS models of C, H and S element is not ideal.(2) Comparing the calibration effect of models based on near infrared spectroscopy, reflectance and transmittance hyperspectral imaging technology, the models which detect calorific value analysis indexes(gross/net calorific value) based on near infrared spectroscopy technique is the optimal models.The optimal pretreatment methods of spectral are: gross calorific value is algorithm of multiple scattering correction, first-order derivative process combined with S-G smooth(MSC+FD+S-G smooth); net calorific value is multiple scattering correction combined with first-order derivative process(MSC+FD). The correlation coefficient of calibration set(Rc) of gross calorific value is 0.904, the mean square root error(RMSEC) is 291 J·g-1, the cross validation correlation coefficient(Rcv) is 0.859, the root mean square error of cross validation(RMSECV) is 348 J·g-1, the correlation coefficient of validation set(Rp) is 0.908, the mean square root error of prediction(RMSEP) is 328 J·g-1, the average deviation(Bias) is 34.616 J·g-1, the relative analysis error(RPD) is 2.22. the established quantitative analysis model of gross calorific value can be used for quantitative analysis; The correlation coefficient of calibration set(Rc) of net calorific value is 0.869, the mean square root error(RMSEC) is 315 J·g-1, the cross validation correlation coefficient(Rcv) is 0.822, the root mean square error of cross validation(RMSECV) is 363 J·g-1, the correlation coefficient of validation set(Rp) is 0.850, the mean square root error of prediction(RMSEP) is 365 J·g-1, the average deviation(Bias) is 34.979 J·g-1, the relative analysis error(RPD) is 1.94, The results show that, the quantitative analysis model could be used to estimate the net calorific value, the accuracy of the model can be further improved.(3) Comparing the calibration effect of models based on near infrared spectroscopy, reflectance and transmittance hyperspectral imaging technology, the models which detect industrial analysis indexes(moisture, volatile matter, ash and fixed carbon) based on hyperspectral reflectance technique is the optimal models.The optimal pretreatment methods of spectral are: ash is algorithm of multiple scattering correction combined with standard processing(MSC+Autoscale); moisture is algorithm of standard normal variate combined with standard processing(SNV+Autoscale); volatile matter is first-order derivative process combined with mean center(FD+Mean Center); fixed carbon is algorithm of standard normal variate combined with mean center(SNV+Mean Center). Determine the competitive adaptive reweighted sampling(CARS) for optimal variable selection algorithm. The quantitative analysis models of industrial analysis indexes have been optimized. The correlation coefficient of validation set(Rp) of model of ash is 0.931, the mean square root error of prediction(RMSEP) decreases as 0.369%, the relative analysis error(RPD) is 8.86. The model of ash performs a good prediction ability, while the prediction effect of the other 3 models of industrial analysis indexes is not ideal, and can not be used for quantitative analysis in practice.