[Study on combinatorial optimization of spectral principal components using successive projections algorithm].

Successive projections algorithm (SPA) was employed to select the optimal combination of principal components (PCs) which were obtained by principal component analysis. Short-wave near infrared spectra of milk powder was firstly analyzed by PCA, and the optimal combination of obtained first eight PCs was determined by SPA. The optimal PC combination of fat content prediction was PC1 , PC2, PC 4, PC5, PC6 and PC7, and the combination for protein content prediction was PC1, PC2, PC3, PC4, PC5 and PC8. Least-squares support vector machine models inputted by different PC combination were established to predict fat and protein content, respectively. Both the fat and protein content prediction results of the PC combination selected by SPA were better than those of first four PCs to first eight PCs. Rp2, and root mean square errors for prediction and residual predictive deviation of prediction results of the PC combination selected by SPA were 0.989, 0.1703 and 9.5343, respectively for fat, and 0.9876, 0.1348 and 8.9274 for protein. The overall results demonstrate that SPA can fast and effectively select the optimal PC combination. The selecting process is simple and does not need abundant parameter debugging.