基于改进免疫遗传算法的近红外光谱变量选择方法  

Selection of Near Infrared Spectral Wavelength Variables Based on Improved Immune Genetic Algorithm

作　　者：陶焕明 高美凤 TAO Huan-ming;GAO Mei-feng(Key Laboratory of Advanced Process Control for Light Industry(Ministry of Education),School of Internet of Things Engineering,Jiangnan University,Wuxi 214122,China)

机构地区：[1]江南大学轻工过程先进控制教育部重点实验室,物联网工程学院,江苏无锡214122

出　　处：《分析测试学报》2021年第10期1482-1488,共7页Journal of Instrumental Analysis

基　　金：国家自然科学基金资助项目(61833007)。

摘　　要：该文在免疫遗传算法(IGA)的基础上,提出一种改进免疫遗传算法(iIGA)用于近红外光谱波长变量的选择。该算法舍去了原算法中固定抗体相似度阈值的思想,取而代之的是抗体相似度阈值自适应,同时引入精英保留策略和贪心算法思想,使得算法朝着正确的方向进行局部性探优。将该算法在玉米的淀粉和蛋白质含量数据集上进行实验测试,建立偏最小二乘(PLS)分析模型,并与IGA、遗传算法(GA)以及全谱方法进行了对比。结果表明,在玉米淀粉含量的预测上,iIGA相较于原IGA算法,预测集均方根误差(RMSEP)从0.3120降至0.2980,预测集预测精度提升4.5%;在玉米蛋白质含量的预测上,RMSEP从0.1244降至0.1103,预测集预测精度提升11.3%。分别对预测淀粉和蛋白质模型的RMSEP值进行显著性检验,F值分别为165.22和182.05,P值分别为9.5×10^(-23)和4.5×10^(-24),P值均小于0.05,因此,iIGA能显著提升模型预测精度。Based on immune genetic algorithm(IGA),an improved immune genetic algorithm(iIGA)was proposed to select the wavelength variables of near infrared spectra.The idea of fixed an⁃tibody similarity threshold in the original algorithm was abandoned in the iIGA,which was replaced with adaptive antibody similarity threshold.Meanwhile,the elitist retention strategy and greedy algo⁃rithm idea were introduced,making the algorithm carry out local optimization in the right direction.The algorithm was tested on the corn starch and protein content data sets to establish a partial least squares(PLS)analysis model,and compared with IGA,genetic algorithm(GA)and full spectrum method.Results showed that the root mean square error of prediction set(RMSEP)of iIGA was re⁃duced from 0.3120 to 0.2980,compared with those of the original IGA algorithm,the prediction accuracy of prediction set was improved by 4.5%.In the prediction of corn protein content,the RM⁃SEP decreased from 0.1244 to 0.1103,the prediction accuracy of prediction set increased by 11.3%.A significant test was carried out for the RMSEP values of starch and protein models,respec⁃tively,in which F values were 165.22 and 182.05,P values were 9.5×10^(-23 )and 4.5×10^(-24),re⁃spectively,and P values were less than 0.05.Therefore,iIGA could significantly improve the pre⁃diction accuracy of the model.

关 键 词：近红外光谱 波长选择 改进的免疫遗传算法 分析模型 预测精度 

分 类 号：O657.3[理学—分析化学]