近红外光谱技术结合4种算法分析尾巨桉-马占相思制浆原料的混合程度与主化学成分  

Analysis of the Mixing Degree and Main Chemical Composition of Eucalyptus Urophylla×Grandis and Acacia Mangium Pulping Materials by Near-infrared Spectroscopy Combined with 4 Algorithms

作　　者：吴珽 梁龙[1] 朱北平[1] 邓拥军[1] 房桂干[1] WU Ting;LIANG Long;ZHU Bei-ping;DENG Yong-jun;FANG Gui-gan(Key Laboratory of Biomass Energy and Material,Key Laboratory of Chemical Engineering of Forest Products,National Forestry and Grassland Administration,National Engineering Laboratory for Biomass Chemical Utilization,Institute of Chemical Industry of Forest Products,Chinese Academy of Forestry,Nanjing 210042,China;Gold East Paper (Jiangsu) Co.,Ltd.,Zhenjiang 212132,China)

机构地区：[1]中国林业科学研究院林产化学工业研究所,生物质化学利用国家工程实验室,国家林业和草原局林产化学工程重点实验室,江苏省生物质能源与材料重点实验室,江苏南京210042 [2]金东纸业(江苏)股份有限公司,江苏镇江212132

出　　处：《分析测试学报》2020年第11期1351-1357,共7页Journal of Instrumental Analysis

基　　金：中国博士后科学基金资助项目(2019M661780);国家重点研发计划项目(2017YFD0601005)。

摘　　要：为缓解我国木浆供应压力,满足混合原料制浆的实际需求,该文进行了近红外光谱快速分析混合制浆原料的研究。采集145个人为控制尾巨桉含量的尾巨桉-马占相思混合样品的近红外光谱,用常规方法测定其综纤维素、聚戊糖、Klason木质素含量。对原始光谱进行一阶导数与标准正态变换预处理后,分别运用偏最小二乘法、支持向量机法、人工神经网络法和LASSO算法建立尾巨桉、综纤维素、聚戊糖、Klason木质素含量分析模型。其中LASSO法建立的尾巨桉和综纤维素含量分析模型最优,预测均方根误差(RMSEP)分别为1.80%、0.60%;绝对偏差(AD)分别为-3.03%~3.17%、-1.03%~0.98%,模型性能可满足较精确的快速分析。偏最小二乘法建立的聚戊糖含量分析模型最优,RMSEP为0.75%,AD为-1.26%~1.33%;支持向量机法建立的Klason木质素含量分析模型最优,RMSEP为0.48%,AD为-0.82%~0.86%,两个模型性能适用于非精确性的分析。该研究为混合制浆原料的快速分析提供了可能,同时也证实了LASSO算法的适用性。In order to alleviate the pressure of wood pulp supply in China and meet the actual demand of pulping with mixed pulpwood,a study was conducted on near-infrared rapid analysis of mixed pulpwood.145 mixed samples of Eucalyptus urophylla×grandis-Acacia mangium were prepared,in which the content of Eucalyptus urophylla×grandis was manually controlled.The near-infrared spectra of these samples were collected,and the contents of holocellulose,pentosan and Klason lignin were analyzed by traditional methods.After the original spectra were pretreated by first derivative and standard normal variate,the analysis models for contents of Eucalyptus urophylla×grandis,holocellulose,pentosan and Klason lignin were established by partial least squares method,support vector machine method,artificial neural network method and LASSO algorithm,respectively.Among them,models for contents of Eucalyptus urophylla×grandis and holocellulose established by LASSO algorithm were the best,with their root mean square error of prediction(RMSEP)values of 1.80%and 0.60%,and their absolute deviation(AD)ranges of-3.03%-3.17%and-1.03%-0.98%,respectively,which could be used for accurate and rapid analysis.Besides,the model for content of pentosan established by partial least squares was the best,with its RMSEP value of 0.75%and an absolute deviation range of-1.26%-1.33%,while the model for content of Klason lignin established by the support vector machine method was the best,with its RMSEP value of 0.48%and an absolute deviation range of-0.82%-0.86%.The performance of the two models was suitable for inaccurate analysis.This study provides the possibility for rapid analysis of mixed pulpwood,and also confirms the applicability of LASSO algorithm.

关 键 词：近红外光谱 LASSO算法 混合原料 制浆造纸 成分含量 

分 类 号：O657.33[理学—分析化学] TQ35[理学—化学]