Adaptive Learning of Hybrid Models for Nonlinear Model Predictive Control of Distillation Columns

Abstract

Abstract Our previous work has shown that replacing parts of the classical compartmentalization model reduction approach for distillation columns by offline-trained artificial neural networks (ANNs) improves computational performance. In real-life applications, the absence of a high-fidelity model for data generation can, however, prevent the deployment of this approach. Therefore, we propose a method that utilizes solely plant measurement data, starting from a small initial data set and then continuously adapting to newly measured data. We demonstrate the approach in closed-loop simulations and compare to benchmarks using either the high-fidelity model or an offline trained reduced model for control.