Online Model Maintenance in Real-time Optimization Methods

Abstract The performance of model-based optimization methods, like Real-time Optimization (RTO), relies on the model accuracy and adequacy. However, features of the process may be unknown and/or the system behavior can drastically change with time (e.g. system degradation). Therefore, even if we have a perfect model in the beginning, we may end up making decisions based on a poor model. This paper proposes a method that adapts the model structure online, based on an available model set, while simultaneously estimates the model parameters. The problem is presented in a superstructure framework and solved using a mixed-integer nonlinear formulation. Then, the updated model is combined with Output Modifier Adaptation, an RTO variant, for economic optimization. Our method is tested in a continuous stirred-tank reactor and a gas lifted oil well network. The results show that we can select the correct model structure, update its parameters and, simultaneously, converge to the plant optimum.