Doctoral seminar

An adaptive shooting approach for dynamic optimization of unstable processes:
In dynamic optimization of chemical engineering problems, often large and partially unstable systems occur. These belong to a particularly challenging class of problems which cannot be tackled by the so-called single shooting method. Single shooting methods are able to solve large-scale dynamic systems efficiently, but lack the ability to cope properly with unstable dynamics. On the contrary, the so-called multiple shooting method is capable of solving these unstable system, however, it lacks the ability of handling large-scale problems efficiently. Yet another approach is to solve the dynamic optimization problem through the formulation of the indirect problem, which is rather complicated and inefficient.

This presentation will introduce an innovative extension of single shooting to the so-called adaptive shooting. The method detects instabilities during computations and reformulates the problem in order to deal with the unstable modes of the system. This approach improves significantly the condition of the non-linear program (NLP) compared to the single shooting method, as only the unstable manifold of the state vector is parameterized. Thus, the small-scale property of the NLP in case of single shooting can be preserved. The method will be demonstrated for a well known chemical engineering benchmark problem.

Speaker: Fady Assassa

Date: Tuesday, September 29, 2009

Time: 16:30

Location: Seminar Room 001, German Research School for Simulation Sciences, Schinkelstr. 2a, 52062 Aachen