汽油辛烷值近红外光谱检测的改进极限学习机建模方法  

Novel modeling method based on improved extreme learning machine algorithm for gasoline octane number detection by near infrared spectroscopy

作　　者：胡碧霞 张红光[1] 卢建刚[1] 鄢悦 李雪园 韩金厚 刘彤 陈金水[1] 孙优贤[1] 

机构地区：[1]浙江大学工业控制技术国家重点实验室,浙江杭州310027

出　　处：《南京理工大学学报》2017年第5期660-665,共6页Journal of Nanjing University of Science and Technology

基　　金：国家自然科学基金(61590925;U1509211)

摘　　要：为提高近红外光谱法检测汽油辛烷值的精度,该文提出一种汽油辛烷值近红外光谱检测的改进极限学习机(i ELM)新型建模方法。该算法融合了极限学习机算法(ELM)与基于变量投影重要性系数的改进叠加偏最小二乘回归(VIP-SPLS)模型算法,有效解决了ELM模型隐含层输出矩阵维数高和高度共线性的问题。采用该算法对汽油辛烷值的近红外光谱检测数据进行建模,发现改进极限学习机模型的精度比现有的偏最小二乘回归模型和极限学习机模型分别提高20.0%和29.3%,验证了方法的有效性。实验表明,该文方法可用于汽油辛烷值的近红外光谱检测,检测精度良好。In order to improve the accuracy of gasoline octane number detection by the near infrared （NIR） spectroscopy, an improved extreme learning machine （ iELM ） algorithm combined with theextreme learning machine（ ELM） algorithm and the improved stacked partial least square regression based on the variable importance in the projection （ VIP-SPLS） algorithm is proposed here. And it solves the problem of high dimension and high collinearity in the output matrix of hidden layer of the ELM algorithm effectively. Then the proposed method is applied to a commonly used benchmark NIR spectral data of gasoline octane number detection. The results show that, compared with the PLS model and the ELM model, the accuracy of iELM model is increased by 20. 0 % and 29. 3 % respectively. The experiment shows that the iELM algorithm can be applied to the gasoline octane number detection by the near infrared spectroscopy and its accuracy is satisfactory.

关 键 词：汽油辛烷值 近红外光谱 模型 极限学习机 偏最小二乘 变量投影重要性系数 

分 类 号：O657.3[理学—分析化学]