(264b) Nonlinear Predictive Control of Integrated Process Systems

The chemical process industry is an intensely competitive environment, in which cost reduction represents a critical factor towards increasing profit margins. The need to lower utility costs and energy consumption, and to improve raw material use has spurred the development and implementation of increasingly integrated process designs that make extensive use of material and energy recycling. While yielding a significant reduction in capital and operating costs, process integration also gives rise to operational and control challenges, including strong interactions between units and nonlinear behavior.

The difficulties posed by the control of integrated plants have been, in numerous instances, successfully addressed within the linear Model Predictive Control (MPC) paradigm. To date, linear MPC remains the approach of choice for regulatory control and plant operation around a steady state, as it permits centralized decision making while accounting for economic optimality, model structure and operating constraints.

The current economic environment is, however, highly dynamic. As a result, optimal plant operation requires frequent switching among different product grades and production rates. The limited range of validity of data-driven process models makes enforcing transitions between significantly different operating points via linear MPC a difficult task. In such circumstances, the use of nonlinear (potentially first-principles) process models in formulating the MPC problem would be preferable. The implementation of the latter approach, referred to as nonlinear Model Predictive Control (NMPC) is, however, hindered by the high dimensionality size and stiffness associated with the nonlinear models of complex integrated processes.

In the present paper, we introduce a novel NMPC framework suitable for transition control in a broad class of integrated process systems which feature significant material and/or energy recycling and material purge streams. Specifically, we rely on our previous results [1] to synthesize a control structure at the process unit level, as well as to derive reduced-order representations of the plant-wide dynamics. We subsequently utilize the resulting models to develop and solve a non-stiff, nonlinear MPC formulation for the entire plant. We also introduce a modification of this two-tiered control structure, which incorporates economic parameters in the formulation of the NMPC optimization problem. Finally, we present a case study concerning a reaction-separation network and demonstrate the excellent robustness and flexibility of the proposed approach (implemented using the NMPC package OptCon [2]) when performing economically optimal transitions between different steady states.

[1] Baldea, M. and Daoutidis, P., Control of integrated process networks?A multi-time scale perspective , Comp. Chem. Eng., 31, 426-444, 2007.

[2] Nagy, Z. K., Mahn, B., Franke, R., and Allgower, F., Efficient output feedback nonlinear model predictive control for temperature control of industrial batch reactors, Control Engineering Practice, 15, 839-859, 2007.