为提高制浆树种的利用效率, 缓解国内制浆造纸原料短缺的现状, 降低行业污染与总体成本, 尝试将近红外光谱技术用于海南省制浆树种的成分含量分析, 以期根据实时所得成分含量相应调整工艺参数。 用结构简单、 易改装的全息光栅分光近红外光谱仪采集了海南省常见的适龄制浆树种(尾细桉、 尾巨桉、 尾叶桉、 马占相思和粗果相思)共205个样本的近红外光谱, 按传统实验室方法分析其主要成分——综纤维素和木质素的含量。 选择合适的预处理方法与偏最小二乘法结合, 建立了两种分析模型, 并通过遗传算法剔除不相关的变量, 筛选出特征波段, 明确综纤维素和木质素的特征吸收, 优化了模型。 其中综纤维素分析模型建立时采用Savitzky-Golay 13点3倍平滑、 矢量归一化和一阶导数预处理原始光谱, 1 150.3~2 362.0 nm波段参与建模。 筛选出的波段包含了1 188~1 196 nm之间CH₃中C—H伸缩振动的二级倍频吸收, 1 742~1 633 nm区间内O—H伸缩振动的一级倍频, 2 112 nm附近O—H变形振动、 O—H伸缩振动的合频等纤维素的特征吸收; 也包含了1 470~1 495 nm之间O—H伸缩振动的一级倍频, 1 906和1 911 nm附近C=O伸缩振动的二级倍频等聚戊糖的特征吸收。 模型RMSEP值为0.55%, 绝对偏差范围为-0.91%~0.87%。 木质素分析模型建立时采用Savitzky-Golay 13点3倍平滑、 多元信号校正和二阶导数预处理原始光谱, 1 137.6~1 872.5和2 131.0~2 424.1 nm波段参与建模。 筛选出的波段包含了1 143 nm附近苯环C—H伸缩振动的二级倍频吸收和CH₃的C—H伸缩振动的二级倍频吸收, 1 670~1 684 nm处苯环C—H伸缩振动的一级倍频, 2 205 nm附近C—H、 C=O伸缩振动的合频等木质素的特征吸收。 模型RMSEP值为0.45%, 绝对偏差范围为-0.76%~0.79%。 两个模型的RPD值分别为4.71和3.47, 均能满足制浆树种主要成分在线快速分析测定的工业需求。 同时, 本研究为制浆树种近红外表征体系的建立提供了理论依据, 对近红外技术助力制浆造纸工业由自动化向智能化转变具有较为显著的意义。

In order to improve the utilization efficiency of pulpwood in Hainan Province, alleviate the shortage of domestic pulping and papermaking materials, and reduce pollution and overall costs in the pulping and papermaking industry, this study aimed to use near-infrared spectroscopy for the analysis of pulpwood. A holographic grating spectroscopic near-infrared spectrometer with a simple structure and easy modification was used to collect the near-infrared spectrum of 205 samples of pulpwood common in Hainan ( E. urophlla× E. tereticornis, Eucalyptus urophylla× grandis, Eucalyptus urophylla, Acacia mangium, Acacia crassicarpa Benth.), and the content of holocellulose and lignin were measured according to the traditional laboratory methods. Suitable pretreatment methods were selected in combination with partial least squares to establish analysis models holocellulose and lignin. Then genetic algorithm was usedto eliminate the irrelevant variables and clarifythe feature absorption of holocellulose and lignin in order to optimize the models. The holocellulose model was established by pretreatment methods of Savitzky-Golay 13 points 3 times smoothing, vector normalization, the first derivative of the original spectrum, with 1 150.3~2 362.0 nm bands participated in modeling. The optimal bands included the characteristic absorption of cellulose such as the 2nd overtone of C—H stretching vibration in CH₃ between 1 188~1 196 nm, the 1st overtone of O—H stretching vibration between 1 742~1 633 nm, the group frequencies of formation and stretching vibration of O—H near 2 112 nm. The optimal bands also included the characteristic absorption of pentosan such as the 1st overtone absorption of O—H stretching vibration between 1 470~1 495 nm, and the 2nd overtone absorption of C=O stretching vibration around 1 906 and 1 911 nm. The RMSEP value of the model was 0.55%, and the absolute deviation range was -0.91%~0.87%. The lignin model was established by pretreatment methods of Savitzky-Golay 13 points 3 times smoothing, MSC, the second derivative of the original spectrum, with 1 137.6~1 872.5 and 2 131.0~2 424.1 nm bands participated in modeling. The optimal bands included the characteristic absorption of lignin such as the 2nd overtone of the C—H stretching vibration in the benzene ring and in the CH₃ near 1 143 nm, the 1st overtone of the C—H stretching vibration in the benzene ring between 1 670~1 684 nm, the group frequencies of stretching vibration of C—H and C=O near 2 205 nm. The RMSEP value of the model was 0.45%, and the absolute deviation range was -0.76%~0.79%. The two models’ RPD values were 4.71 and 3.47, respectively, which can meet the actual needs of online quick analysis and measurement of the main components of pulpwood. At the same time, this study provides a theoretical basis for the establishment of a near-infrared characterization system for pulpwood, and has a significant significance for the near-infrared technology to help the pulping and papermaking industry to change from automation to intelligence.