基于改进堆叠自编码器结合LightGBM的近红外光谱回归算法研究  

Study on Near Infrared Spectral Regression Analysis Method Based on Improved Stacked Autoencoder Combined with LightGBM

作　　者：吴继忠[1] 时艺丹 黄慧[2] 厉小润[3] WU Ji-zhong;SHI Yi-dan;HUANG Hui;LI Xiao-run(Technology Center,China Tobacco Zhejiang Industrial Co.,Ltd,Hangzhou 310008,China;Ocean College,Zhejiang University,Zhoushan 316000,China;College of Electrical Engineering,Zhejiang University,Hangzhou 310027,China)

机构地区：[1]浙江中烟工业有限责任公司技术中心,浙江杭州310008 [2]浙江大学海洋学院,浙江舟山316000 [3]浙江大学电气工程学院,浙江杭州310027

出　　处：《分析测试学报》2023年第9期1112-1118,共7页Journal of Instrumental Analysis

基　　金：国家自然科学基金资助项目(62171404);浙江中烟科技项目(ZJZY2021A020);浙江大学-浙江中烟联合实验室科技项目(2021-KYY-510012-0001)。

摘　　要：为实现高效的近红外光谱非线性回归分析,提出了一种基于改进堆叠自编码器结合LightGBM的近红外光谱回归分析算法。该算法由堆叠自编码器模块与LightGBM模块构成,将堆叠自编码模块得到的隐层特征输入LightGBM模块进行回归分析,通过递进式策略自适应确定堆叠自编码器模块的结构,并利用Optu⁃na框架自动优化LightGBM模块的超参数。为验证方法的有效性,以烟草的还原糖、氯、钾、总氮4种成分为研究对象,利用1911个烟草样本进行建模,并与其他4种近红外光谱回归分析算法进行了对比。经实验验证,烟草还原糖、氯、钾、总氮预测模型的平均R_(P)、RMSEP、R_(P)^(2)分别为0.9110、0.0568、0.8328,预测精度在5种方法中综合最优。在训练集表现相当的前提下,所建方法的预测集精度相较于XGBoost提高1%~40%,过拟合问题得到改善。改进的堆叠自编码器结合LightGBM算法应用于近红外光谱分析表现出良好的成分回归分析能力,可用于烟叶化学成分预测模型的构建。To achieve efficient nonlinear regression analysis of near-infrared spectra,a near-infrared spectrum regression analysis algorithm based on an improved stacked autoencoder combined with LightGBM is proposed.The algorithm consists of the stacked autoencoder module and the LightGBM module.The hidden layer features obtained from the stacked autoencoder module are sent to the LightGBM module for regression analysis.The structure of the stacked autoencoder module is adap⁃tively determined through a progressive strategy,and the hyperparameters of the LightGBM module are automatically optimized using the Optuna framework.To validate the effectiveness of the method,the reducing sugar,chlorine,potassium,and total nitrogen of tobacco were selected as the research objects,and a model was built using 1911 tobacco samples.The results were compared with four other near-infrared spectral regression analysis algorithms.Experimental results showed that the aver⁃age Rp,RMSEP,R_(P)^(2)of the predicted models for tobacco reducing sugar,chlorine,potassium,and total nitrogen were 0.9110,0.0568 and 0.8328,respectively,and the prediction accuracy of the proposed method was the best among the five methods.Given equivalent performance on the training set,the proposed method achieves 1%-40%higher prediction accuracy on the test set compared to XGBoost.The improved stacked autoencoder combined with LightGBM demonstrates good component regression analysis ability in near-infrared spectral analysis and can be used to construct chemical composition prediction models for tobacco.

关 键 词：近红外光谱 回归分析 改进堆叠自编码器 LightGBM 

分 类 号：O657.3[理学—分析化学] TB9[理学—化学]