模拟退火算法(simulated annealing algorithm,SAA)是一种随机搜索、全局优化算法,为提高近红外光谱检测面粉品质模型的准确度与稳健性,实验提出基于SAA优化波长,再结合偏最小二乘(partial least squares, PLS)法建模预测的定量模型,并对SAA中冷却进度表参数设置进行对比分析。实验依据面粉中灰分含量梯度,随机选取126份样本的近红外光谱建立SAA-PLS模型。结果发现,SAA从2 074个波数优选出70个波数,结合PLS建立的定量模型相关系数为0.976 0,交互验证均方根误差(root mean square error of cross validation,RMSECV)为0.022,预测均方根误差(root mean square error of prediction,RMSEP)为0.030 1,全谱建立的PLS模型相关系数为0.778 5,RMSECV为0.066 6,RMSEP为0.076 8。结果表明,基于SAA优化特征谱区,建立灰分定量模型是可行的,且准确度与稳健性明显优于全谱定量分析模型。
英文摘要:
Simulated annealing algorithm (SAA) is a random search algorithm for global optimization. In order to improve the accuracy and robustness of near-infrared spectroscopy (NIR) in detecting wheat flour quality, this paper proposed a quantitative prediction model using global optimization based on SAA combined with partial least squares (PLS). In this algorithm, a comparative analysis was made in different parameter settings of cooling schedule. According to the ash content gradients in flour, the NIR spectra of 126 samples were selected randomly to establish an SAA-PLS model. Results showed that 70 wave numbers were picked out of 2 074 wave numbers using SAA. The quantitative model established using partial least squares exhibited a correlation coefficient (CC) of 0.976 0, a root mean square error of cross validation (RMSECV) of 0.022, and a root mean square error of prediction (RMSEP) of 0.030 1, while the CC, RMSECV and RMSEP values of the PLS model based on the full wave spectra was 0.778 5, 0.066 6 and 0.076 8, respectively. These results indicated that it was feasible to establish a quantitative model for predicting ash content using wavelength optimization based on SAA, which was superior in accuracy and robustness to the full-spectrum model.