近红外光谱结合Lasso算法测定制浆材抽出物含量

The Prediction of Pulpwood Extractives Content by Near Infrared Spectroscopy Combining with Lasso Algorithm

中文关键词:  Lasso算法  近红外光谱  制浆材  抽出物含量

Key Words:Lasso algorithm  near-infrared spectroscopy  pulpwood  extractive content

中文摘要:

      为实现制浆材材性的快速测定,首先用常规方法测定了144个制浆材样品的冷水、热水、苯-醇和1% NaOH抽出物含量,并采集了样品的近红外光谱,然后对原始光谱进行预处理,并运用Lasso算法及交互验证建立最优校正模型。对模型进行独立验证,决定系数R2val分别为0.9186、0.9085、0.9241、0.9760,预测均方根误差分别为0.24%、0.30%、0.28%、0.38%,相对分析误差分别为3.50、3.31、3.63、6.45,绝对偏差分别为-0.42%～0.37%、-0.43%～0.41%、-0.47%～0.40%、-0.55%～0.57%。这些模型预测性能能够满足制浆造纸工业的要求,同时,也证实了Lasso算法用于制浆材抽出物测定的可行性。

Abstract:

      The contents of cold water, hot water, benzene ethanol and 1.0% NaOH extractive of 144 pulpwood samples were analyzed using the traditional methods, meanwhile their near-infrared (NIR) spectra were also collected. After the pretreatment of original spectra, the optimal prediction models were established by using Lasso algorithm and cross-validation. The independent verification of the optimal prediction models showed the coefficients of determination (R2) were 0.9186, 0.9085, 0.9241 and 0.9760. The root mean square error of prediction (RMSEP) were 0.24%, 0.30%, 0.28% and 0.38%. The relative percent deviation (RPD) were 3.50, 3.31, 3.63 and 6.45. The absolute deviation (AD) were -0.42%～0.37%,-0.43%～0.41%,-0.47%～0.40%,-0.55%～0.57% respectively for cold water, hot water, benzene ethanol and 1.0% NaOH extractives. The prediction performance of the four models could meet the need of pulping and paper making industry and meanwhile Lasso algorithm was feasible for the prediction and analysis of pulpwood extractive content.