近红外光谱结合特征变量筛选方法用于农药乳油中毒死蜱含量的测定  

Determination of Chlorpyrifos in Pesticide Formulations Using NIR Spectroscopy and Variable Selection Methods

作　　者：吴瑞梅[1] 王晓[1] 郭平[2] 艾施荣[3] 严霖元[1] 刘木华[1] 

机构地区：[1]江西农业大学工学院,江西南昌330045 [2]江西出入境检验检疫局技术中心,江西南昌330002 [3]江西农业大学软件学院,江西南昌330045

出　　处：《分析测试学报》2013年第11期1359-1363,共5页Journal of Instrumental Analysis

基　　金：国家自然科学基金项目(31271612);江西省自然基金项目(20122BAB204020);江西省科技攻关项目(2011BDH80010);江西省教育厅科技项目(GJJ13272)

摘　　要：为提高毒死蜱农药乳油中有效成分近红外光谱定量分析模型的精度和稳定性。采用联合区间偏最小二乘法(siPLS)结合遗传算法(GA)筛选特征变量,由交互验证法确定最佳主成分因子数及筛选的变量数。结果表明,从全光谱区优选出81个变量,主成分因子数为11时,能建立性能最优的模型,模型预测集的决定系数R_p^2为0.972,预测均方根误差(RMSEP)为0.353%。研究表明,利用siPLS结合GA方法优选特征变量,能大幅度地消除农药乳油光谱变量间的冗余信息和无关信息,降低模型的复杂度,提高农药有效成分预测模型的精度及稳定性。To improve the precision and robustness of the model for the chlorpyrifos active ingredient in pesticide EC by near infrared spectroscopy,synergy interval PLS（siPLS） combined with genetic algorithm（GA） was implemented to optimize the feature variables,and cross-validation method was used to select the optimal PLS factors and the variables.The results showed that the optimal model was achieved with R2p of 0.972,root mean square error of prediction（RMSEP） of 0.353% in the prediction set when 81 variables and 11 PLS factors were included.Experimental results showed that siPLS combined with GA could eliminate a large margin of the redundant information and irrelevant information in pesticide EC spectroscopy,and reduce the complexity of the developed model.The precision and robustness of the model were also improved.

关 键 词：近红外光谱 联合区间偏最小二乘法(siPLS) 遗传算法(GA) 农药制剂 毒死蜱 

分 类 号：O657.3[理学—分析化学] S482.3[理学—化学]