基于局部最小二乘支持向量机的光谱定量分析

Spectral Quantitative Analysis Based on Local Least Square Support Vector Machine Regression

提出了一种基于局部最小二乘支持向量机(LSSVM)的回归方法,以克服待测参数和光谱数据间的非线性。本方法首先通过欧式距离选取局部训练样本子集,然后利用该子集建立LSSVM校正模型。由于每个测试样本建模时要选取不同的训练样本,因此提出相对距离的概念用来改进高斯核函数,使LSSVM的参数对于不同的训练样本具有自调整功能。针对一批汽油样本的实验结果表明,本方法的预测精度优于常见的局部线性建模方法和全局建模方法。

In order to deal with the nonlinearity between the property and spectra of samples in spectral quantitative analysis, a local regression method based on least square support vector machine （LSSVM） was proposed. For a testing sample, local training samples were firstly selected from the total training set by Euclidian distance, and then a local LSSVM calibration model was built to predict the property of the testing sample. Because the selected training samples were different for every regression, the concept of relative distance was introduced to improve Gauss kernel function in order to adapt to different sample sets. Applying the proposed method to a set of gasoline samples, experimental results show its root mean square error of prediction is less than both local linear regression and other conventionally used global regression.