基于近红外数据预测烟草淀粉含量的GA-SVR模型

Application of GA-SVR to NIR parameters choose

近红外数据主成分降维后的因子筛选是建模的关键问题,一般是按照主成分贡献率的大小依次选择主成分因子。本文利用遗传算法进行主成分降维后的因子选择,然后利用支持向量回归方法建立烟叶中淀粉含量的定量预报模型。计算结果表明:用遗传算法和支持向量回归方法联合建模,对于训练集和独立测试集所得淀粉含量计算值与实验室之间的相关系数平方分别为0.91和0.90,结果优于利用主成分贡献率筛选变量后再用支持向量回归建模的结果。

It is very important for a model to choose suitable factors among the PCA variables available from near-infrared spectroscopy (NIRS). The subset of PCA factors is often selected based on PCA contribution rates. In this work, the genetic algorithm was applied to choose PCA factors from NIR parameters, then the support vector regress method was used to construct a model to predict the starch contents with data set of 187 tobacco samples. The correlation coefficient of square (R2) are 0.91 and 0.9 for training set and independent test set respectively. Therefore, the GA-SVR is a useful tool to develop the quantitative model for NIR parameters based on choosing PCA factors using genetic algorithm.