Identification of Hammerstein-Weiner models for nonlinear MPC from infrequent measurements in batch processes

Abstract

Here we introduce a new approach for the identification of Hammerstein-Wiener (H–W) models from infrequent measurements in batch processes. Because concentration measurements during each batch run are very infrequent, such models cannot be effectively estimated directly from the infrequent measurements. This difficulty is bypassed by first developing a dynamic response surface methodology (DRSM) model. One can also calculate the optimal trajectories using this DRSM model and identify the local H-W model around such optimal or any other operational trajectory for control purposes. This is obtained by frequently “sampling” the DRSM model around these trajectories. We demonstrate the efficacy of the proposed approach using the simulation of an isothermal semi-batch reactor with nonlinear reaction mechanism. The MPC controller using the nonlinear H-W model is able to outperform the same type of controller using a linear recursive model identified directly from the original infrequent measurements.