Research On Detection Method Of TFAs Content In Oil Based On NIR

Oil is essential to human body,and its quality and safety are particularly important.In recent years,the concern about this problem that TFAs content in oil exceeds the standard.Trans fatty acids are often produced when high quality oils are deodorized.At present,the detection method of TFAs has some problems such as slow detection speed and complex early processing,which cannot meet the requirements of real-time nondestructive online detection.Near infrared spectroscopy technology is convenient and efficient,which is suitable for the quality monitoring in the production process.Therefore,this thesis presents a fast method for detecting trans fatty acids(TFAs)content in soybean oil based on near infrared spectroscopy.In order to control the content of trans fatty acids(TFAs)in the process of deodorization.100 soybean oil samples with different TFAs content were prepared.The TFAs content of the samples was determined by GC as the standard value,and then the spectra of soybean oil samples were scanned by near-infrared spectrometer,and then the spectral data were denoised by different methods.After comparative analysis,it was found that the denoising effect of MSC was the best.In order to explore the absorption characteristics of TFAs in the near-infrared region,multiple iPLS method are used to select the characteristic band of the spectral data,and the characteristic absorption band of TFAs is selected as 7258-7443/6502-6691/6120-6309cm-1.On this basis,Kalman filtering algorithm is used to select the characteristic wavelength variables,and 27 TFAs characteristic wavelength variables are optimized.In the training of DBN model,it is found that the performance of DBN model is best when the number of hidden layers is 3 and the number of hidden layer nodes is50-35-90.Finally,the DBN model with this parameter setting was used to establish the regression model of TFAs content,which is compared with the regression model of trans fatty acid content established by PLS,The results show that:The whole spectrum after noise reduction is modeled,the prediction effect of DBN model is better than that of PLS model,the Coefficient of Determination(R~2)is 0.8794,Root-Mean-Square Error of Prediction(RMSEP)is 0.0603 and Relative Standard Deviation(RSD)is 2.18%;The spectral data after feature band selection is modeled,the prediction effect of PLS model is better than that of DBN model;27selected characteristic wavelength variables are modeled,and DBN has a good prediction effect,the Coefficient of Determination(R~2)is 0.9584,Root-Mean-Square Error of Prediction(RMSEP)is 0.0350 and Relative Standard Deviation(RSD)is1.31%,It indicates that DBN model has better generalization ability,and a small amount of wavelength variables can be used to achieve better prediction results,which can meet the actual detection needs,and provide technical support for the realization of online detection and regulation of TFAs content in the process of oil processing and the production of low/zero TFAs oil products.