Rapid determination of the key temperatures in diesel distillation process based on near-infrared spectroscopy

Abstract

The distillation temperature of petroleum is the significant information for the determination of refinery operating conditions. As the standard laboratory test method, ASTM D86 is often cost, time-consuming and not well suitable for on-line determination. In this paper, we proposed a simple approach to the prediction the key temperatures in diesel distillation process based on the raw near infrared (NIR) spectra of samples. After the NIR spectrum was decomposed by discrete wavelet transform to get the different NIR sub-signals, the selected sub-signals by genetic algorithm (GA) were superposed to form the new effective signal. Then stepwise regression was employed to build the linear prediction models. The proposed strategy was applied to predict the five distillation temperatures of diesel simultaneously, and the obtained Rp2 values of independent external test set were more than 0.96 as well as the average relative errors (ARE) were lower than 1%, which showed that the predicted values were well correlated with the reference values. Compared with the other conventional methods, such as PLS, iPLS, siPLS and stepwise, the proposed approach could obtain more accurate and reliable prediction models. This study not only indicated the validity of the new approach, but also provided an important support for the further realization of on-line NIR detection to predict the distillation temperatures.