Prediction Of Solubility Of Vegetable Oils In Supercritical CO₂ By The Quantitative Structure-Property Relationship Model

As an effective and green process technology,the supercritical CO₂ extraction technology has been widely used in extracting natural products from the plant.It has many excellent physicochemical properties such as non-toxic and harmless,non-flammable and low cost,etc.Besides,supercritical CO₂ avoids introducing impurities into solutes,which brings convenience particularly to the extraction and separation of lipid-like natural products such as edible oils.As is known to us,vegetable oils consist mainly of triglycerides,and solubility of vegetable oils in supercritical CO₂ is a very important physicochemical parameter for studies including phase equilibrium,the extraction technology and the manufacture and design of apparatuses.However,it is difficult and time-consuming that solubility data of solutes in supercritical CO₂ are experimentally measured which can cause larger errors.So it has always been the difficult point and the hot topic of relevant studies that solubility of triglycerides and vegetable oils in supercritical CO₂ is predicted using the equation of state（EOS）,semi-empirical and those theoretical models based on the quantum chemistry and the stoichiometric chemistry combining with the computer science.Therefore,the main contents in this paper are as following:In this study,solubility data of triglycerides in supercritical CO₂ were collected from references and a new model based on quantitative structure-property relationship（QSPR）was developed to predict these solubility data of triglycerides in supercritical CO₂.And solubility of four vegetable oils including the soybean oil,the sunflower oil,the rapeseed oil and the green coffee bean oil in supercritical CO₂ were predicted using the developed QSPR model after simplifying triglyceride components in vegetable oils.Besides,solubility data of the soybean oil in supercritical CO₂ were experimentally measured by both static and dynamic method and were compared with those that were estimated by the model and were taken from references.The main contents and conclusions are as following:Ⅰ.The previous papers were reviewed in which solubility of solutes in supercritical CO₂ was predicted by theoretical,semi-empirical and recently popular neural network models,and some advantages and shortcomings existing in these models were introduced and suggested that the support vector machine（SVM）model was a more appropriate method to predict solubility of solutes in supercritical CO₂.Ⅱ.The solubility data of pure triglycerides（purity>99%）were collected from references,and a new SVM model based on the genetic algorithm（GA）was established to predict solubility of pure triglycerides in supercritical CO₂.The results indicate that correlation coefficients（R）of the training set and the test set are 0.986and 0.982,respectively,and corresponding root mean standard errors（RMSE）are14.10%and 19.30%,respectively.After outliers were removed,values of R of the training set and the test set increase to 0.992 and 0.984,respectively,and corresponding RMSE decrease to 10.70%and 11.70%,respectively.In addition,internal validations,the external validation and the sensitivity analysis were conducted to evaluate the stability of the model and the contribution of every variable on solubility to further illustrate that the model is suitable for predicting triglyceride solubility in supercritical CO_2.Ⅲ.Two methods,the pseudo-binary components of CO₂ and triglyceride and the triglyceride component were used to simplify vegetable oils as a type of pure triglyceride and the detailed triglyceride structures,respectively in turn the problem to estimate solubility of complex vegetable oils in supercritical CO₂ is changed into estimate a pure triglyceride solubility or some pure triglyceride solubility by the developed GA-SVM model.The solubility of the soybean oil,the sunflower oil,the rapeseed oil and the green coffee bean oil were predicted and were used to compare with their own ones at the same conditions from references.The results indicate that the performance of the latter are always superior to that of the former,and the average absolute relative deviations（AARD）of the latter is around 30%when temperature is below 333.15K and pressure is below 60MPa,which is lower than the maximum of40%according to the reference.The solubility of the soybean oil in supercritical CO₂ were measured by the static-dynamic method at the conditions of 313.15K-353.15K and 20MPa-60MPa and the obtained experiment data were used to compare with experimental ones taken from references and those estimated by the above model method.The results indicate the AARD is 33.41%when solubility data from references was measured by the GA-SVM model coupling with the triglyceride component method,while the AARD is 40.15%when the same model method was used to predict our solubility data measured by the experiment.The solubility of soybean oil in supercritical CO₂ can be estimated by the GA-SVM model combining with the method of simplifying vegetable oils according to the fact two AARD are near the AARD maximum of 40%.