Predictive Model Of Group-method For Diesel Oil Properties Based On Near-Infrared Spectroscopy

As a rapid, non-destructive, safe and environment-protective analytical technique, near-infrared (NIR) spectroscopy analytical technique has got broad application in the field of petrochemical industry. Since traditional laboratory analytical methods for diesel oil products properties are usually high cost and long time delayed, this thesis gives a research on the group-method predictive model for diesel oil products properties based on the NIR spectroscopy.In this thesis, the superiority of NIR spectroscopy technique on the prediction of petroleum products group structure as well as the relationship between C-H groups of NIR spectroscopy and diesel oil products properties are discussed, and an idea for directly predicting the diesel oil products properties with crucial C-H groups based on the NIR spectroscopy is proposed. Firstly, the NIR spectroscopy data for diesel oil is pretreated, then according to the relative theory of group absorbance and diesel oil properties,several crucial groups are chosen as modeling input variables and a non-linear diesel oil properties predictive model is developed , whose parameters are determined by Gauss-Newton method. The character of the model is that the computing amount depends on number of input variable samples rather than the input dimension and can get high model performance while treating non-linear problems. Using the NIR spectroscopy data of diesel oil samples from Eigenvector Research Corporation, NIR spectroscopy predictive models for diesel oil properties, such as cetane number, have been established by using PLS, BP-ANN, SVM and the group-method separately. The simulation experiments show that the group-method model gains higher predictive accuracy and better generalization than the other models, because of mechanism relationship between the cetane number and the NIR absorbency information in the group-method.