2025 AIChE Annual Meeting

(377g) Control Strategies with States Estimation for Slurry-Phase Reactors for Heavy Oil Hydroprocessing

Authors

Luis Ricardez-Sandoval, University of Waterloo
Jorge Ancheyta, ESIQIE, Instituto Politécnico Nacional
As the industry faces the challenge of reducing greenhouse gas emissions, various alternatives are being explored to minimize the use of fuel oils. However, during this transition period, the demand for these fuels remains consistent. In recent years, light hydrocarbon reserves have diminished, increasing the need to process non-conventional oils, particularly heavy crude oils. Several technologies exist for heavy oil upgrading, among which hydrotreating has gained attention over the past decades. Slurry-phase reactors have emerged as a promising technology for heavy oil hydrocracking and hydrotreating, offering high conversion rates. However, due to their relatively recent development, few mathematical models accurately describe their operation. Many existing models rely on empirical correlations derived under conditions different from those of hydrotreating, introducing significant model uncertainties.

This work explores the implementation of an Extended Kalman Filter (EKF) to handle process uncertainties and integrates it with a Model Predictive Control (MPC) strategy. The MPC approach is compared against a classical PI controller for heavy oil hydrocracking in a slurry-phase reactor under process and measurement uncertainties. The reactor model incorporates axial dispersion effects and a kinetic scheme for both distillation lumps (vacuum residue, vacuum gasoil, middle distillates, and naphtha) and SARA fractions (saturates, aromatics, resins, and asphaltenes) in the presence of a mineral catalyst. The system consists of 39 state variables, which are estimated in real-time using an EKF. These estimated states serve as initial conditions for the MPC algorithm.

To replicate industrial conditions, only temperature measurements are assumed to be available, while the remaining state variables—representing lump compositions—are inferred via model-based optimization. Uncertainties in the system arise from plant-model mismatch and measurement noise. A comparative analysis is conducted between a PI controller and an MPC strategy, both leveraging an EKF for state estimation based only on temperature observations. Simulation results demonstrate that uncertainties significantly degrade the performance of the PI controller, whereas the MPC-EKF framework exhibits superior robustness, effectively maintaining the set-point under input and set-point disturbances while simultaneously minimizing energy consumption.