RON predicted of gasoline by NIR based on ICA and SVM

Petroleum and its products are complex mixture, and how to precise analysis its components is an important part of the oil industry. In this paper, we proposed a components forecasting methods for gasoline octane value prediction based on independent component analysis (ICA) and support vector machine (SVM). By evaluating the accuracy of the models with two feature optimization methods(principal component analysis, PCA and successive projections algorithm, SPA) and two prediction models (neural network, ANN and minimum mean square squares, PLS). The results show that: the prediction relative error of the ICA-SVM model is only 0.34% and the squared correlation coefficient(R2) reached 0.99583, 0.97891, and the mean squared error(MSE) was 0.0010276, 0.013122 on the training set and test set respectively which are better than other compound models. This method in this paper has positive significance for the oil component analysis.