Inferring Light-cycle-oil Stream Properties Using Soft Sensors

J. Joucowski, P. M. Ndiaye, M. L. Corazza and M. K. Lenzi

Abstract
The intensive necessity of hydrotreatment units for diesel production is pushing petroleum companies to seek alternatives to frame the produced streams into ultra low sulphur diesel (ULSD) specifications. One of the main difficulties in ULSD production is the presence of compounds from dibenzothiophenes (DBT), which are of difficult hydrotreatment. The LCO cutpoint control represents an interesting alternative to overcome this situation. Thus, the objective of this work was to develop a soft sensor using linear models and neural networks considering a set of historical data of temperature, pressure and flow obtained from industrial plant information. Lab and process data concerning a period of 18 months was successfully used to infer 10%, 30%, 50%, 70% and 90% ASTM D-86 recovery temperature. Based on correlation matrix plots, using lab data as the dependent variable and plant data as independent variable, different models were developed for LCO cutpoint prediction. For all models, correlation coefficient between model predictions and experimental data were above 0.95.

This work is licensed under a Creative Commons Attribution 4.0 International License

Keywords
Inference, soft sensor, diesel, property, neural network