Quantitative Study Of Edible Blend Oils By UV-Vis Spectroscopy Combined With Chemometrics

Blend oil is a type of product that has been deeply processed by vegetable oils.Since blend oil can balance the fatty acids in vegetable oils,it is very important for human nutrition and health.However,there has been no unified standard and specification for the blend ratio of the blend oil for a long time,which has led to the ambiguity of the types and contents of specific components in the blend oil.Therefore,it is necessary to develop a rapid quantitative detection method to ensure the quality of the blend oil.In this thesis,the feasibility of quantitative analysis of blend oil using UVvis spectroscopy combined with chemometrics was investigated.The specific research contents are as follows.1.A weighted multi-scale regression method based on empirical mode decomposition(EMD)is proposed for quantitative analysis of binary and ternary blend oils.Binary blend oil samples composed of soybean oil and peanut oil and ternary blend oil samples composed of soybean oil,peanut oil,and sesame oil were obtained according to a certain mass percentage.Then the UV-Vis spectra of binary and ternary blend oil samples were collected.Using the adaptability of EMD,the spectra of binary and ternary blend oil samples were decomposed respectively,and a series of intrinsic mode functions(IMFs)and a residual(r)were obtained,then the maximum decomposition mode was determined by optimizing the number of decomposition.Based on the maximum decomposition mode,the EMD-SVR model of the peanut oil component of binary blend oil and the sesame oil component of ternary blend oil were established respectively.In order to verify the modeling effect of EMD-SVR,the prediction results of EMD-SVR were compared with PLS and SVR.The results showed that in the binary blend oil samples,the RMSEP values of PLS,SVR and EMD-SVR were 9.7339,5.7716 and 3.7554,respectively,and the R values were 0.9461,0.9879 and 0.9933,respectively.RMSEP of ternary blend oil samples were 0.5676,0.6050,0.4999,and R were 0.9845,0.9803,0.9866,respectively.By comparison,it can be concluded that EMD-SVR has the lowest RMSEP and the highest R in both binary and ternary blend oil.Therefore,the prediction accuracy of the model can be effectively improved by adaptive decomposition of the original spectra.2.A grey wolf algorithm-partial least squares(GWO-PLS)modeling method based on swarm intelligence is proposed for quantitative analysis of quaternary blend oils.First,102 quaternary blend oil samples were obtained by blending soybean oil,sunflower oil,peanut oil and sesame oil according to a certain mass percentage.Then,the UV-Vis spectra of quaternary blend oil were collected.GWO was introduced into UV-Vis spectra to select variables by using its strong local searching ability and solving precision in solving optimization problems.The influences of the number of iterations and the number of wolves on the optimization results were investigated,and then the GWO-PLS quantitative model is established on the basis of the optimal number of iterations and the number of wolves.In order to illustrate the modeling effect of GWO after variable selection,the prediction results of PLS and GWO-PLS were compared.The results showed that after variable selection,the RMSEP of soybean oil,sunflower oil,peanut oil and sesame oil decreased by 44%,49%,63%,45%,and the R value increased by 0.31%,9.47%,9.91%,17.56%,respectively.At this time,the number of variables retained by soybean oil,sunflower oil,peanut oil and sesame oil was 55,35,50,66,respectively.Therefore,the establishment of PLS model after GWO variable selection can improve the prediction performance of the model.