化工学报 ›› 2013, Vol. 64 ›› Issue (12): 4410-4415.DOI: 10.3969/j.issn.0438-1157.2013.12.021

• 过程系统工程 • 上一篇    下一篇

一种新的近红外光谱信息区间选择方法

徐龙1, 卢建刚1, 杨秦敏1, 陈金水1, 施英姿2   

  1. 1. 浙江大学控制科学与工程学系, 工业控制技术国家重点实验室, 浙江 杭州 310027;
    2. 杭州师范大学教育学院, 浙江 杭州 311121
  • 收稿日期:2013-08-12 修回日期:2013-08-25 出版日期:2013-12-05 发布日期:2013-12-05
  • 通讯作者: 卢建刚
  • 作者简介:徐龙(1989- ),男,硕士研究生。
  • 基金资助:

    国家自然科学基金项目(21076179);国家重点基础研究发展计划项目(2012CB720500)。

A new near-infrared spectroscopy informative interval selection method

XU Long1, LU Jiangang1, YANG Qinmin1, CHEN Jinshui1, SHI Yingzi2   

  1. 1. State Key Laboratory of Industrial Control Technology, Department of Control Science and Engineering, Zhejiang University, Hangzhou 310027, Zhejiang, China;
    2. College of Education, Hangzhou Normal University, Hangzhou 311121, Zhejiang, China
  • Received:2013-08-12 Revised:2013-08-25 Online:2013-12-05 Published:2013-12-05
  • Supported by:

    supported by the National Natural Science Foundation of China (21076179) and the National Basic Research Program of China (2012CB720500).

摘要: 基于间隔策略的信息波长选择是近红外光谱分析中广泛应用的一种方法。针对传统算法忽略非线性因素的缺点,首次考虑将最小二乘支持向量机(least-squares support vector machine,LSSVM)方法应用于间隔选择策略,进而提出了一种新的波长选择方法iLSSVM(interval LSSVM)。该算法克服了传统间隔选择算法依赖于线性模型的缺陷,对存在较强非线性的光谱数据能够准确地选择最优信息区间,极大地减少建模变量并显著改善模型预测精度。应用两组业界标准的光谱数据来验证该算法的性能,并和传统方法进行了比较。实验结果表明,在两组数据集上该算法取得的标准预测偏差(root mean square error of prediction,RMSEP)分别比全谱PLS建模降低了20%和4%,比传统的间隔偏最小二乘算法(interval partial least-squares,iPLS)降低了28%和2%。

关键词: 近红外光谱, 间隔选择, 非线性模型, 最小二乘支持向量机

Abstract: Strategy based on interval selection is widely used in the near-infrared spectroscopy analysis. Inspired by interval partial least-squares method (iPLS),the present paper proposed a new wavelength method combining interval selection strategy with least-squares support vector machine (LSSVM).By overcoming the shortcomings of traditional interval selection methods whose predictive ability totally depend on the linear model,this new algorithm,named as iLSSVM (interval LSSVM),can select the optimal informative interval more reasonably to significantly improve the model prediction accuracy with less modeling variables.Two real near-infrared datasets were applied to this new approach and the prediction performance was compared to the other interval selection methods.The experimental results demonstrated that the root mean square error of prediction (RMSEP) of this new method is 20% and 4% smaller than that of full-spectrum PLS modeling method respectively,and is 28% and 2% smaller than that of the traditional iPLS (interval partial least-squares) method respectively.

Key words: near-infrared spectroscopy, interval selection, nonlinear model, LSSVM

中图分类号: