论文部分内容阅读
相思树是一种速生纸浆材,苯醇抽提物含量对木材的制浆得率有一定影响。近红外光谱分析技术能对木材的化学成分含量进行低成本快速检测。多模型方法是一种预测效果好且易于掌握的近红外光谱分析建模方法,已被用于建立相思树、毛白杨和欧美杨某些化学成分含量的近红外光谱预测模型,取得较好的建模效果。先用多模型方法建立了相思树苯醇抽提物含量和Klason木质素含量的近红外光谱分析模型。结果表明Klason木质素含量的预测效果优于苯醇抽提物含量的预测效果。然后在多模型方法的基础上,用预测误差较小的Klason木质素含量优化构建了苯醇抽提物含量的预测模型,使苯醇抽提物含量的预测效果得到改进。模型的拟合优度从0.792 8提升到0.827 1,预测值与实验值之间的相关系数从0.907 4提升到0.922 5。不同于已有的多模型方法,在优化建模时并不要求所使用的两种化学成分含量之间具有近似线性关系。还对优化构建的苯醇抽提物含量预测模型,通过减少每个子模型中待定常数的个数,增强了模型的稳定性,进一步改进了模型的预测效果。随着这方面研究工作的增多,未来该建模方法有希望应用于某些预测效果一般的化学成分含量,使这些化学成分含量的近红外光谱分析效果得到改进。
Acacia is a fast-growing pulp material, the alcohol extract content of wood pulp yield has a certain impact. Near-infrared spectroscopy enables the rapid detection of the chemical content of wood at a low cost. The multi-model method is a predictive and easy-to-grasp near-infrared spectroscopy modeling method that has been used to establish near-infrared spectral prediction models for the chemical constituents of Acacia, Populus tomentosa and Populus euphratica and to obtain better Modeling effect. The multi-model method was used to establish the near-infrared spectral analysis model of the content of alcohol extract and Klason lignin in Acacia. The results show that the predictive effect of Klason lignin content is better than that of the alcohol extract content prediction. Then, based on the multi-model method, the predictive model of the content of benzene-alcohol extract was optimized by optimizing the Klason lignin content with small prediction error, and the prediction of the content of benzene-alcohol extract was improved. The goodness of fit of the model was improved from 0.792 8 to 0.827 1, and the correlation coefficient between the predicted value and the experimental value increased from 0.907 4 to 0.922 5. Unlike existing multi-model methods, it is not necessary to have an approximately linear relationship between the two chemical compositional contents used in optimization modeling. The model of predicting the content of benzene alcohol extract was also optimized, and the stability of the model was enhanced by reducing the number of undetermined constants in each submodel. The prediction effect of the model was further improved. With the increase of research work in this field, the modeling method is expected to be applied to some chemical compositions with predicted effects in the future, and the NIR analysis results of these chemical compositions are improved.