Prediction of octane number loss based on Kernel-PCA and BP-MLP

Gasoline is an important fuel resource, which is obtained from the heavy oil in crude oil by catalytic cracking. Octane number is the most important index of gasoline combustion performance. In the refining process of gasoline, not only the sulfur and olefin will decrease, but also the octane number will be lost, resulting in huge economic loss. Therefore, it has important research value to analyze and optimize chemical process model and reduce octane number loss. In this experiment, 325 data samples collected from FCC gasoline refining unit were used to obtain feature vector combinations of 367 operating variables after dimensionality reduction by Kernel PCA algorithm. Factor analysis method in SPSS software was used to carry out original feature analysis on these linear combinations. According to the contribution of each variable to each combination, the main modeling variables were selected, and finally the model was established to predict octane number loss by using BP neural network algorithm.