近红外光谱结合改进鲸鱼算法优化模型BAS-WOA-SVR检测藤椒油掺伪  

Detection of Adulteration of Vine Pepper Oil by Near-Infrared Spectroscopy Combined With Improved Whale Optimization Algorithm Model BAS-WOA-SVR

作　　者：许素安 王家祥 刘勇 XU Su-an;WANG Jia-xiang;LIU Yong(College of Mechanical and Electrical Engineering,China Jiliang University,Hangzhou 310018,China)

机构地区：[1]中国计量大学机电工程学院,浙江杭州310018

出　　处：《光谱学与光谱分析》2023年第2期569-576,共8页Spectroscopy and Spectral Analysis

基　　金：国家自然科学基金项目(51105348)资助。

摘　　要：鉴于藤椒油市场良莠不齐,以近红外光谱技术为基础,藤椒油为研究对象,展开对藤椒油掺伪检测研究。首先将纯藤椒油作为基底油,按比例配置掺入大豆油、玉米油、葵花籽油得到油样,采集藤椒油掺伪样品的近红外光谱数据;光谱数据经归一化处理后采用标准正态变换(SNV)、多元散射矫正(MSC)进行预处理,然后采用竞争性自适应重加权算法(CARS)、连续投影算法(SPA)进行特征数据提取,组合不同预处理算法与特征数据提取算法,通过支持向量机回归(SVR)建立藤椒油掺伪预测模型。结果表明:MSC-CARS-SVR模型校正集和预测集的决定系数(R^(2))最高,校正集R^(2)达到了0.7561,预测集R^(2)达到0.7052;均方根误差(RMSE)最小,校正集RMSE达到0.743,预测集RMSE达到0.794。为了提高模型的准确性,采用鲸鱼算法(WOA)和改进鲸鱼算法(BAS-WOA)优化SVR模型,改进的鲸鱼算法以每一次鲸鱼群的最优鲸鱼作为当前天牛须的出发位置,分别探索左右须前进,计算前进后的目标函数,如果目标函数优于当前最优鲸鱼的值,则用前进后的天牛位置替换鲸鱼位置,进而实现了天牛须算子对鲸鱼算法的改进。用WOA优化SVR模型,相比之下精度最高的为MSC-CARS-WOA-SVR模型,校正集R^(2)达到0.8591,预测集R^(2)达到0.8216;校正集RMSE降低到了0.374,预测集RMSE降低到0.495。相比于传统的SVR模型精度和性能都有较明显提升。用BAS-WOA优化SVR模型,精度最高的是MSC-CARS-BAS-WOA-SVR模型,校正集R^(2)高达0.9551,预测集R^(2)高达0.9439;校正集RMSE降低到了0.054,预测集RMSE降低到0.081。相比于WOA优化算法,BAS-WOA优化的模型精确度和性能都有了进一步提升,模型预测集R^(2)从0.8216提高到0.9439,预测集RMSE从0.495降低为0.081。鲸鱼算法优化SVR模型容易陷入局部极值和收敛速度问题,改进的鲸鱼算法通过天牛须算法的左右须探寻来改进鲸鱼算法不足,从而提升算法的全�Due to the uneven market of rattan pepper oil,based on near-infrared spectroscopy technology,rattan pepper oil is the research object,and the research on the adulteration detection of rattan pepper oil is carried out.First,pure rattan pepper oil was used as the base oil,and the adulterated soybean oil,corn oil,and sunflower oil were prepared in proportion to obtain oil samples.The near-infrared spectroscopy was used to collect the spectral data of the oil samples to obtain the adulterated near-infrared spectral data of rattan pepper oil.The spectral data are normalized and preprocessed by Standard Normal Variation(SNV)and MultivariativeScatter Correction(MSC).And then,the feature data is processed by Competitive Adaptive Reweighting Sampling(CARS)and SuccessiveProjection Algorithm(SPA).Extraction,combining different preprocessing algorithms and feature data extraction algorithms,and establishing a prediction model of vine pepper oil adulteration through Support Vector Machine regression(SVR).The results show that the coefficient of determination(R^(2))of the calibration set and prediction set of the MSC-CARS-SVR model is the highest,the calibration set R^(2)reaches 0.7561,and the prediction set R^(2)reaches 0.7052;the root mean square error(RMSE)is the smallest,and the calibration set RMSE reaches 0.743,The prediction set RMSE reaches 0.794.In order to improve the accuracy of the model,the Whale Optimization Algorithm(WOA)and the Improved Whale Optimization Algorithm(BAS-WOA)are used to optimize the SVR model.The left and right beards are moved forward,and the objective function after the advance is calculated.If the objective function is better than the current optimal whale value,the position of the whale is replaced by the position of the beetle after the move forward,thereby realizing the improvement of the beetle operator on the whale algorithm.When WOA optimizes the SVR model,compared with the MSC-CARS-WOA-SVR model with the highest accuracy,the R^(2)of the calibration set can reach 0.8591,and the R^(2)of t

关 键 词：近红外光谱 藤椒油 改进鲸鱼算法(BAS-WOA) 支持向量机回归(SVR) 掺伪 

分 类 号：TP274[自动化与计算机技术—检测技术与自动化装置]