Research On Modeling Methods For Detecting Content Of Trans Fatty Acids In Soybean Oil Based On Terahertz Spectroscopy

Edible oil is an essential component for human diet,and its quality and safety are crucial to human health.In recent years,the problem of excessive trans-fatty acids in edible oil is particularly serious,which has aroused the comprehensive attention of social circles.At present,the existing methods for detecting content of trans-fatty acids have some problems,such as lots of chemicals,slow detection speed,complex detection procedures,complicated sample pretreatment and so on.It is difficult to meet the requirements of modern society for detecting oil quality that should be rapid,accurate,simple and on-site.Terahertz spectroscopy analysis can overcome many disadvantages of traditional methods,it is more suited to quality control in the process of oil production.Therefore,this paper presents the method based on terahertz spectroscopy technology to detect the content of trans-fatty acids in edible oil.Taking soybean oil for example that is most widely used in China,the focuses of the study is mainly on the data processing and modeling methods for detecting trans-fatty acids content by terahertz spectrum analysis.First,34 soybean oil samples with different content of trans-fatty acids were prepared,the trans-fatty acids contents were precisely detected by gas chromatography,meanwhile the terahertz time-domain spectra of samples were collected and transformed into frequency domain spectra by Fourier transform,then the terahertz absorption spectra and reflection spectra were obtained by calculating the optical parameters.On the basis of analyzing the characteristics of the terahertz absorption spectra of soybean oils,one abnormal sample was removed.Second,the 33 samples were ordered according to the content of trans-fatty acids and divided,28 samples were used as training set for establishing the correction models,the rest of 5 samples were used as prediction set for validating the models.In order to find the best model,the three algorithms were used to build models respectively and comparatively analyzed,which are partial least squares(PLS),support vector machine regression(SVR)and least squares support vector machine regression(LS-SVR).Finally,the LS-SVR method was found to be the best one,the root mean square error of prediction(RMSEP)is 0.3246,the coefficient of determination(R2)is 0.9792,and the relative standard deviation(RSD)is 3.81%,respectively,which can completely meet the practical demand of detection.In order to further improve the accuracy of prediction,the three algorithms were used respectively to optimize the model parameters of LS-SVR and compared,which are grid search method,particle swarm optimization(PSO)algorithm and genetic algorithm(GA).It is showed that PSO algorithm had better effect on optimizing the parameters of LS-SVR and the results were more stable,the root mean square error of prediction(RMSEP)is 0.0763,the coefficient of determination(R2)is 0.9989,and the relative standard deviation(RSD)is 0.90%,respectively,the prediction accuracy of the model is enhanced significantly.The research of this paper demonstrated the feasibility of detecting the content of trans-fatty acids in edible oils by using terahertz spectroscopy,which laid theoretical foundation for developing special instruments of terahertz spectroscopy and realizing on-site monitoring of edible oil quality.