Nonlinear Model Predictive Control Using Decoupled A-B Net Formulation for Carbon Capture Systems - Comparison with Algorithmic Differentiation Approach

Abstract

A novel state-space-based multivariable constrained nonlinear model predictive control (NMPC) strategy is described. The NMPC strategy uses enhanced Decoupled A-B Net (DABNet) based dynamic reduced models (D-RMs) as core predictive models. The control objective cost-function and its analytical derivatives, explicitly utilize the decoupled nature of the DABNet model, which provides computational advantages over automatic differentiation (AD) methodologies. The output-node parallelization strategy described in this paper is demonstrated to provide comparable computational advantages as that found in linear multi-rate MPC algorithms, while furnishing superior disturbance rejection and setpoint tracking characteristics of the NMPC. The approach has been presented and compared on a multi-time-scale solid-sorbent-based bubbling fluidized bed (BFB) carbon capture process.