Abstract: To obtain the viscosity index of VGO rapidly, a prediction model was established by random forest regression algorithm, based on the near infrared spectroscopy and viscosity index data of 70 representative VGO samples. Based on the importance measurement of each feature in the random forest algorithm, the recursive feature elimination method was used to select wavelength variables in NIR. The more robust model was built by selecting 10 characteristic wavelengths as the input features for the model and determining the hyper parameters (the number of trees in the forest nt of 150, the number of features to consider when splitting nv of 5) by 10-fold cross validation. The prediction standard deviation of 7 new samples is 2.28 with R2 of 0.98, indicating high accuracy and strong generalization ability of this model.

Key words: vacuum gas oil, viscosity index, prediction, near infrared spectroscopy, recursive feature elimination method, random forest algorithm