基于偏最小二乘变换与Kriging模型的全局优化方法

针对传统Kriging模型在多变量(高维)输入全局优化中因超参数过多而引发收敛速度慢,精度低,建模效率不高问题,提出了基于偏最小二乘变换技术和Kriging模型的有效全局优化方法.首先,构造偏最小二乘高斯核函数;其次,借助差分进化算法寻找满足期望改进准则最大化条件的新样本点;然后,将不同核函数和期望改进准则组合,构建四种有效全局优化算法并进行比较;最后,数值算例结果表明,基于偏最小二乘变换的Kriging全局优化方法在解决高维全局优化问题方面相比于标准的全局优化算法在收敛精度及收敛速度方面更具优势.

Carob Origin Classification by FTIR Spectroscopy and Chemometrics

The goal of this study was to use Fourier transform mid-infrared （FTIR） spectroscopy for discrimination of samples of pods and seeds of carob from three Moroccan regions. The origin of samples Pods and seeds of carob could be distinguished from their IR spectra and this measurement was used for discriminate analysis. A multivariate analysis procedure based on the combined use of Hierarchical Cluster Aanalysis （HCA） and Partial Least Squares-Discriminant Analysis （PLS-DA） was tested and provided good classification results. Three distinctive clusters were recognised, related to the three Moroccan regions. Afterwards, PLS-DA was used for the discrimination and classification of the origin of the various Pods and seeds of carob samples. The results demonstrated that the combined use of FTIR and chemometric analysis （cluster analysis and discrimination by PLS- DA） can be used to rapidly and simply determine the origin of carob pulpe samples.