人工神经网络结合近红外光谱测定制浆材材性  

Prediction of Pulpwood Properties by Artificial Neural Network Coupling with Near Infrared Spectroscopy

作　　者：吴珽 房桂干[1] 梁龙[1] 张新民[2] 赵振义[2] 

机构地区：[1]中国林业科学研究院林产化学工业研究所江苏省生物质能源与材料重点实验室国家林业局林产化学工程重点开放性实验室生物质化学利用国家工程实验室,江苏南京210042 [2]华夏科创仪器有限公司,北京100008

出　　处：《纸和造纸》2015年第8期83-86,共4页Paper and Paper Making

基　　金：国家林业局948项目"农林剩余物制机械浆节能和减量技术引进"(2014-4-31)

摘　　要：用常规方法测定了141个制浆材样品的综纤维素、木素和苯醇抽出物含量并采集了样品的近红外光谱。对原始光谱进行多元散射校正后,运用反向传输神经网络结合交互验证的方法,确定模型参数并建立样品综纤维素、木素、苯醇抽出物的校正模型。独立验证中模型的决定系数Rval2分别为0.9478、0.9724、0.9367;预测均方根误差（RMSEP）分别为0.61%、0.46%、0.24%;相对分析误差（RPD）值分别为4.38、6.02、3.97;绝对偏差（AD）分别为-1.16%~0.93%、-0.67%~0.81%、-0.37%~0.47%、预测均方根误差和绝对偏差基本符合对误差的要求,3个校正模型能够满足制浆造纸工业中制浆材材性的快速测定。A study was conducted to use the traditional method to analyze the composition of holocellulose, lignin and benzene -alcohol extract of 141 pulpwood samples and the near-infrared （NIR） spectra were also collected. Back propagation artificial neural network（BP-ANN） method and cross-validation were used to confirm factor and build the calibration models for holocellulose, lignin and benzene-alcohol extract after the original spectra were pretreated by multipliplicative scatter correction（MSC）, The independent verification of the calibration models showed the coefficient of determination （R2val） were 0.9478, 0.9724, 0.9367, respectively, The root mean square error of prediction （RMSEP） were 0.61%, 0.46%, 0.24%, respectively.The relative percent deviation （RPD） were 4.38, 6.02, 3.97, respectively. And the absolute deviation （AD） were -1.16%~0.93%, -0.67%-0.81%, -0.37%-0.47%, respectively, qhe root mean square error of prediction and the absolute deviation basically met the needs of error and the three calibration models could realize the rapid determination in pulping and paper making industry for the good predictive performance.

关 键 词：人工神经网络 近红外光谱 制浆材 材性 

分 类 号：O657.33[理学—分析化学] TS77[理学—化学]