Real Time Optimization of the Gasoline Blending Process with Unscented Kalman Filter

Abstract

Gasoline blending is a critical process in petroleum refineries. Real-time optimization (RTO) techniques have been popular with the applications for the blending process for optimization purpose. However the dependency of RTO on the measurement of the component impairs its applicability. Therefore how to utilize the blending model and the product measurement to free RTO from the component measurement is the major research topic in this paper. Unscented Kalman Filter, due to its ability to estimate the parameter for nonlinear model, is chosen to estimate component properties based on the product measurement. The RTO strategy is then proposed with the UKF method for the recipe calculation periodically. Furthermore, the proposed RTO is tested with the gasoline blending benchmark problem, while the results are compared with the ideal blending case. The accuracy of the component estimation and the efficiency of the RTO are verified with the results.