Adaptive discretization methods for the efficient solution of dynamic optimization problems
Flexible production in the process industry requires rapid changes of load and product specifications. Dynamic optimization provides a suitable mathematical framework for finding optimized operating policies. This thesis presents improvements of numerical techniques for dynamic optimization problems using a direct sequential solution approach, employing adaptation concepts. A one-step extrapolation method for sensitivity integration is shown being advantageous over commonly applied multi-step integrators. Further, a wavelet-based adaptive refinement algorithm for efficient control variable discretization is introduced. A method for automatic detection of the control switching structure is presented subsequently. It supports the physical interpretability of the numerical solution and leads to a highly accurate, yet efficient control parameterization. Finally, some extensions to on-line applications are shown. All concepts are illustrated by means of various case studies, including semi-batch and continuous process applications, as well as a large-scale industrial process example.