A multi‐scenario nonlinear model predictive control approach for robust product transitions

Abstract

Dynamic product transitions are ubiquitous operations in the processing industry. When a first‐principles dynamic model is deployed for real system representation, the calculation of the dynamic optimal trajectory for product transition can be cast as an optimal control problem. A common practice in addressing the solution of optimal product transitions lies in the assumption of free of uncertainty first‐principle models. Ignoring the effect of model uncertainty on product transitions can result in unfeasible dynamic trajectories. In this work, an optimization scenario approach, featuring variable scenario weighting functions, is deployed for assessing the impact of model uncertainty on the control actions such that feasible and optimal transition trajectories are computed featuring minimum deviation from target values. The optimization approach was applied to three nonlinear reaction systems. The results demonstrate that when the variable weighting optimization scenario approach is suitable for approximating model uncertainty, feasible transition trajectories can be calculated at relatively low computational cost (for small or medium scale systems).

A weighted local steady‐state determination approach based on the globally optimal economic steady‐states

“…Therefore, in this paper the coaching capability of the GOESS is fully utilized to coordinate the local economic optimality and the fastest dynamic response. The GOESS can be obtained using Equation (6), upon which the closest local steady-states (CLSS) can be evaluated using Equation (18):…”

“…Model predictive control (MPC) has been one of the most popular advanced process control (APC) strategies used in industrial fields. [1][2][3][4][5][6] Among them, two-layer model predictive control (TLMPC) architecture is extensively used in industrial control software, such as DMCplus and DMC3 of AspenTech. [7] Morshedi et al and Brosilow et al proposed a steadystate optimization (SSO) method based on linear programming, which established the basic architecture of TLMPC.…”

“…Model predictive control (MPC) has been one of the most popular advanced process control (APC) strategies used in industrial fields. [ 1–6 ] Among them, two‐layer model predictive control (TLMPC) architecture is extensively used in industrial control software, such as DMCplus and DMC3 of AspenTech. [ 7 ]…”

Robust Nonlinear Model Predictive Control for Two-Stage Anaerobic Digesters

“…In min–max approaches, a closed-loop min–max formulation calculates control inputs that minimize the worst-case cost with respect to the defined bounded set of parameters. , Tube-based MPC ensures that the closed-loop trajectories lie in a tube that satisfies the constraints using an ancillary controller . In the multistage NMPC, uncertainties are modeled using a scenario tree, and the future control inputs are optimized. , Recently, to address the concerns of prediction uncertainty, robust NMPC has focused on providing a mathematical framework for modeling uncertainty values by an uncertainty set and optimizing problems based on the worst scenario. ,,− The basic ideas in this approach were introduced by Huang et al and Huang and Biegler, where different scenarios characterize the uncertainties, and the computed control action is feasible over the entire uncertainty region.…”