论文部分内容阅读
近红外(NIR)定量分析通常涉及多个组分,采用遗传算法和自适应建模策略,建立了能够对多组分同时定量的多目标最小二乘支持向量机(LS-SVM),并将其应用于玉米中四个组分和连翘中两个活性成分的NIR分析。结果表明多目标遗传算法配合自适应建模策略可保证优化收敛于全局最优解。所建玉米多目标LS-SVM模型明显优于PLS1和PLS2模型;连翘多目标LS-SVM模型与PLS模型均可取得较好的校正和预测效果。两组数据中,径向基神经网络(RBFNN)模型均出现过拟合现象。多目标LS-SVM和单目标LS-SVM性能相近,但多目标LS-SVM建模运行一次即可得到结果,在NIR多组分定量分析中具有潜在应用优势。
Near-infrared (NIR) quantitative analysis usually involves multiple components. Genetic algorithms and adaptive modeling strategies are used to establish a multi-objective least squares support vector machine (LS-SVM) It is applied to NIR analysis of two active ingredients in four components of corn and forsythia. The results show that the multi-objective genetic algorithm with adaptive modeling strategy can ensure optimal convergence to the global optimal solution. The multi-objective LS-SVM model of maize was significantly superior to PLS1 and PLS2 models. The multi-target LS-SVM model and PLS model of Forsythia suspensa can achieve better results of calibration and prediction. In both sets of data, RBFNN models all over-fitted. The performance of multi-target LS-SVM and single-target LS-SVM are similar, but multi-target LS-SVM model can be run once to get the result, which has potential application advantages in NIR multi-component quantitative analysis.