稀疏主成分分析-随机森林算法结合荧光光谱的土壤表面油种识别  

Sparse principal component analysis-random forest algorithm combined withoptimized spectroscopyfor identification of soil surface oil species

作　　者：贺兆南 景敏[1,2] 韩亨通 刘盼 计丰 陈曼龙 HE Zhaonan;JING Min;HAN Hengtong;LIU Pan;JI Feng;CHEN Manlong(School of Mechanical Engineering,Shaanxi University of Technology,Hanzhong 723000,China;Shaanxi Provincial Key Laboratory of Industrial Automation,Hanzhong 723000,China)

机构地区：[1]陕西理工大学机械工程学院,汉中723000 [2]陕西省工业自动化重点实验室,汉中723000

出　　处：《分析试验室》2025年第2期216-222,共7页Chinese Journal of Analysis Laboratory

基　　金：陕西省重点产业创新链项目(2021ZDLSF06-07);陕西省自然科学基础研究项目(2022JM-383);陕西理工大学人才启动项目(SLGRCQD2103)资助。

摘　　要：为解决光谱法鉴别土壤表面石油污染物种类中的光谱重叠问题,提出将稀疏主成分分析(SPCA)与随机森林(RF)算法相结合进行油种识别的方法。实验设计了针对土壤表面润滑油的荧光光谱采集系统,以不同种类的发动机油、齿轮油和摩托车油作为研究对象,从含有不同浓度润滑油的土壤样品中提取荧光光谱,并以全波段点、主成分分析(PCA)和SPCA进行特征提取后的光谱作为RF、梯度提升决策树(GBDT)和决策树(Decision Tree)的输入量,润滑油的种类作为输出量。结果表明,经过SPCA后,3种模型的分类准确率分别为98.8%, 97.7%和92.2%;与全波段相比,准确率分别提高了5.5%, 5.5%和6.7%;与PCA相比,准确率分别提高了1.1%, 3.3%和5.6%。该方法实现了对土壤表面石油污染物种类较高效且稳定的分类精度,表明设计的数据采集系统与SPCA-RF算法相结合,可用于土壤表面石油种类鉴别。To solve the spectral overlapproblem in the process of identifying oil pollutant species on the soil surface using spectral analysis,a method of combining sparse principal component analysis(SPCA)with random forest(RF)algorithm for oil species identification was proposed.The fluorescence spectraacquisition system of lubricating oil on soil surface was designed,taking different kinds of engine oil,gear oil,and motorcycle oil as research objects,extracting fluorescence spectra from soil samples containing different concentrations of lubricating oils,and using the spectra after feature extraction with full-band points,principal component analysis(PCA)and SPCA methods as the inputs of RF,gradient boosting decision tree(GBDT),and decision tree,and the kinds of lubricating oils as the output quantity.The results showed that the classification accuracies of the three models after SPCA were 98.8%,97.7%,and 92.2%,respectively,which were 5.5%,5.5%,and 6.7%higher than that of the full-band,and 1.1%,3.3%,and 5.6%higher than that of PCA,respectively.A more efficient and stable classification accuracy of soil surface petroleum contaminant species was achieved,indicating that combining the designed data acquisition system with the SPCA-RF algorithm can be used for the identification of soil surface petroleum species.

关 键 词：荧光光谱 润滑油 种类判别 稀疏主成分分析 

分 类 号：O657.3[理学—分析化学]