Real-time refinery optimization with reduced-order fluidized catalytic cracker model and surrogate-based trust region filter method

Abstract

Since its initial development in the 1980′s, Real-Time Optimization (RTO) has been widely appreciated as an efficient way to optimize process decision variables and improve economic performance of refineries. RTOs consist of nonlinear optimization models with hundreds of thousands of equations, which are built within equation-oriented (EO) modeling platforms. With increasing size and complexity of RTO applications, there is increased demand for improved optimization strategies. To address this demand, surrogate models for complex refinery units have been embedded within the general EO framework for RTO. Moreover, the recent trust region filter (TRF) optimization strategy allows great flexibility in the choice of surrogates, while ensuring convergence to the optimum of the rigorous RTO model. This study considers this approach for a real-world refinery. The Petrobras S.A. RECAP unit in Mauá, Brazil runs an RTO refinery model with an Aspen RTO optimizer to maximize the profit within two hour cycles. To reduce the computational burden, we embed a reduced model (RM) to replace the detailed (truth) model for the residue fluid catalytic cracking (RFCC) unit, and implement a TRF optimization strategy. The TRF driver is written in Python and integrates with the RFCC truth model, the Aspen-EO RECAP model, and the Aspen RTO optimizer. The approach is illustrated on three real-world scenarios in order to demonstrate the effectiveness and efficiency of this RM-based optimization strategy.