利用马拉硫磷在紫外/可见光波长范围内的不同浓度梯度的吸光度光谱数据, 建立其快速有效的定量预测分析模型。 在预测模型的建立过程中, 参与建模的波长变量和校正集样本的优劣对定量分析模型的预测能力有着决定性作用。 首先对实验样本是否存在异常样本进行检查, 然后将200.08~750.04 nm波长范围的光谱数据采用不同预处理方法并建立PLS模型, 进而将预处理结果最优(均值中心化)的光谱数据采用竞争性自适应重加权采样(CARS)算法和蒙特卡洛无消息变量消除法(MC-UVE)分别筛选出关键波长变量并建立相应的PLS预测模型, 模型结果表明, CARS算法在关键变量筛选的性能上优于MC-UVE算法; 再将CARS算法筛选出的18个波长变量(为原来变量数的1.137 8%)结合Kennard-Stone (K-S)算法和蒙特卡洛交叉验证(MCCV)分别优选出的44个建模样本(原来样本数的88%)建立CARS-K-Ss-PLS和CARS-CCVs-PLS定量预测模型, R2p分别为0.998 2和0.998 9, RMSEP分别为0.863 4和1.026 2, RPD分别为24.163 5和20.330 1, CARS-K-Ss-PLS模型略优于CARS-CCVs-PLS模型。 结果表明, CARS算法能够淘汰与样本浓度相关性较弱的变量, 同时有效剔除无关光谱信息, K-S算法能帮助选择更优的建模样本集, 马拉硫磷农药的紫外/可见光吸收光谱结合CARS算法和K-S算法所建立的CARS-K-Ss-PLS模型能够用来定量预测马拉硫磷农药浓度。 研究工作为利用光谱技术快速检测有机磷农药浓度提供一定的理论依据和实验基础, 在有机磷农药快速检测领域具有良好的应用前景。

In this study, the fast and effective quantitative prediction analysis model was established by using the absorption spectrum data of different concentration gradients of malathion in the ultraviolet/visible wavelength range. In the process of establishing a prediction model, the quality of the calibration set samples and wavelength variables involved in the modeling plays a decisive role in the predictive ability of the quantitative analysis model. Therefore, firstly checked whether there were abnormal samples in the experimental samples, then used the different preprocessing methods for the spectral data in the wavelength range of 200.08 to 750.04 nm and then established corresponding PLS model, Further based on the spectral data of the optimal preprocessing result (mean centering), competitive adaptive weighted algorithm (CARS) and Monte Carlo-uninformative variable elimination method (MC-UVE) were used to select the key wavelength variables respectively and established corresponding PLSprediction model. Model results indicated that CARS algorithm was superior to MC-UVE algorithm in the performance of key variable screening; then 18 wavelength variables (1.137 8% of the original variable number) selected by CARS algorithm combined with the 44 modeled samples (88% of the original sample number) respectively obtained from Kennard-Stone (K-S) algorithm method and Monte Carlo cross-validation method (MCCV) to establish CARS-K-Ss-PLS and CARS-CCVs-PLS quantitative prediction model, which R2p were 0.998 2 and 0.998 9, RMSEP were 0.863 4 and 1.026 2, and RPD were 24.163 5 and 20.330 1, as a result the CARS-K-Ss-PLS model was slightly better CARS-CCVs-PLS model. The experimental results showed that the CARS algorithm could eliminate variables with weak correlation with sample concentration and effectively eliminate irrelevant spectral information. The K-S algorithm can help to select a better modeling sample set. UV-Vis absorption spectrum of malathion pesticides combined with the CARS-K-Ss-PLS model established by the CARS algorithm and K-S algorithm can predict malathion pesticide concentration. This study provides a certain of the important theoretical basis and experimental basis for the rapid detection of organophosphorus pesticide concentration by spectroscopy technology, and has a good application prospect in the field of rapid detection of organophosphorus pesticide.