Real-time economic optimization for an integrated plant via a dynamic optimization scheme

Conventional real-time optimizers are based on steady-state models and their effectiveness on plants with long-lived dynamics are thus limited. This is particularly true for tightly integrated plants with material recycle loops and other mass/energy integration loops, which tend to show distinct time-scale separation in their dynamic behavior. The use of steady state model limits the execution frequency of the RTO and precludes the utilization of dynamic degrees of freedom, ultimately leading to suboptimal results. Researchers have suggested to combine unit-level controls and plant-wide economic optimization into a single dynamic optimization but the demand for modeling accuracy and computation may be too high for such an approach to be feasible in practice. We propose a two-layer architecture for dynamic plant-wide optimization and control, in which the upper layer performs a dynamic optimization of the integrated plant to determine economically optimal setpoints for the lower layer performing control functions at the unit level. To alleviate the unrealistic modeling and computational requirements, we propose the plant-wide dynamic optimization at a rate significantly lower than those of the controllers. Slow-scale plant-wide models are less "stiff" and therefore thought to be more robust to model errors. We discuss how to obtain a "slow"-scale plant-wide model for a chosen optimization frequency and the interfacing of the slow-running plant-wide dynamic optimizer with the fast running unit controllers. An example is given to compare the various approaches.