Model reduction for dynamic real-time optimization of chemical processes

The value of models in process industries becomes apparent in practice and literature where numerous successful applications are reported. Process models are being used for optimal plant design, simulation studies, for off-line and online process optimization. For online optimization applications the computational load is a limiting factor. The focus of this thesis is on nonlinear model approximation techniques aiming at reduction of computational load of a dynamic real-time optimization problem. Two types of model approximation methods were selected from literature and assessed within a dynamic optimization case study: model reduction by projection and physics-based model reduction. Model order reduction by projection is partially successful. Even with a strongly reduced number of transformed differential equations it is possible to compute acceptable approximate solutions. Projection does not provide predictable results in terms of simulation error and stability and does not reduce the computational load of simulation. On the other hand, physics-based model reduction appeared to be very successful in reducing the computational load of the sequential dynamic optimization problem.
Design, Engineering and Production