An Approach to Deal with Non-Convex Models in Real-Time Optimization with Modifier Adaptation

MAXIMILIANO GARCÍA; JUAN PABLO RUIZ; MARTA BASUALDO

12th International Symposium on Process Systems Engineering and 25th European Symposium on Computer Aided Process Engineering

Elsevier

Lugar: Copenhagen; Año: 2015; p. 899 - 904

Real Time Optimization (RTO) uses a model of the process to calculate the optimaloperating conditions of the plant. In order to deal with plant-model mismatches,measurement-based adaptation strategies are used. The modifier-adaptation methodperforms corrections in the cost and constraint functions in the model of theoptimization problem to match the necessary optimality conditions for the plant. Bydoing this, an operating point that satisfies the KKT optimality conditions of the actualprocess is obtained. This methodology ensures convergence to the (local) optimum of the plant. However, complex processes usually correspond to models with a large number of non-linearities and non-convexities, often leading to multiple local solutions. Typical algorithms that are used to solve these problems cannot guarantee the global optimum; hence, non-convex models are not properly exploited. This paper aims at dealing with this situation by incorporating a global optimization methodology within the modifier-adaption framework. In order to implement this composite structure we used GAMS to handle the global optimization problem and Matlab to handle the modifier-adaptation scheme. The advantage of this new technique is shown through anillustrative case study.