Globally optimal dynamic real time optimization without model mismatch between optimization and control layer

Abstract

Global optimality of dynamic operations ensures maximum economic benefits over time. We introduce Dynamic Real Time Optimization (D-RTO) method which uses the same model for D-RTO and for Nonlinear Model Predictive Controller (NMPC), thereby eliminating the model mismatch between the D-RTO and the NMPC. An integration framework based on two different predictive horizons for time-scale decomposition is proposed. Fast updates and reoptimization of the dynamic trajectories ensure that the dynamic operation is optimal over time. Proposed global optimization algorithm is a variation of normalized multi-parametric disaggregation (NMDT) where NMDT is modified by the use of a bivariate partitioning to solve the D-RTO problem. Elimination of the mismatch between D-RTO and NMPC models significantly reduces the effort required to eliminate the model errors between D-RTO, NMPC and the plant. Three chemical process examples are included to illustrate the proposed framework.