| Popis: |
The catalytic cracking gasoline refining process is accompanied by the loss of octane, and the traditional chemical process modeling does not respond to the process optimization in a timely manner. In this paper, an improved least partial squares algorithm is proposed to predict the octane loss. According to the idea of dimensionality reduction before modeling, firstly, based on multivariate statistical significance analysis, data mining techniques are used to screen the data samples for modeling main variables and analyze their rationality. Secondly, the traditional partial least squares prediction model was built by extracting the partial least squares components of the modeled main variables, and the R2 of this prediction model was found to be only 0.324528, the prediction effect is not good. Finally, based on the model, we improved the model by introducing the intersection and square terms, and calculated the R2 of the improved model to be 0.673542. The results showed that the improved model nearly doubled the prediction accuracy compared with the traditional partial least squares prediction model, and was able to reasonably predict the octane loss in the FCC gasoline refining process. |
