Publica
Hier finden Sie wissenschaftliche Publikationen aus den FraunhoferInstituten. Decision Support by Multicriteria Optimization in Chemical Production
 Küfer, KarlHeinz; Ruzika, Stefan; Halffmann, Pascal ; Deutsche Gesellschaft für Operations Research DGOR: Multikriterielle Optimierung und Entscheidungsunterstützung : Tagungsband GOR Entscheidungstheorie und praxis 2018, Kaiserslautern Wiesbaden: Springer Gabler, 2019 ISBN: 9783658270407 ISBN: 3658270403 ISBN: 9783658270414 DOI: 10.1007/9783658270414 S.6175 
 Gesellschaft für OperationsResearch, Arbeitsgruppe Entscheidungstheorie und praxis (GOR Annual Meeting) <2018, Kaiserslautern> 

 Englisch 
 Konferenzbeitrag 
 Fraunhofer ITWM () 
Abstract
Modelbased process engineering makes it possible to obtain insight into design alternatives. Since the number of alternatives can be huge and the evaluation criteria to sort out the most relevant alternatives are conflicting, identifying the most favourable ones can be a formidable task. In this work, it is demonstrated that adaptive Pareto optimization techniques can be used to address this task efficiently and reliably. Once a sufficient number of Pareto points that represents the Pareto boundary to a required accuracy is found, an interactive decision support scheme allows to explore correlations and tradeoffs between the different quality criteria. This work describes the introduction and realization of this scheme in industrially relevant contexts. Examples for significant process improvements that were achieved are given.
Furthermore, since mathematical optimization requires many model evaluations, the challenge of shaping the model structure such that it meshes with the optimization requirements is addressed. Therefore, structural reformulations of the equilibrium stage model and asymptotic simplifications are proposed. Additionally, a hybrid modelling approach is presented in which physical models are augmented with statistical elements in order to obtain models of entire chemical processes suitable for optimization.
Intimately related to datadriven model approaches is the task to estimate reliability of model predictions and, if necessary, to plan new experiments in order to obtain additional data. The optimal design of experiments is put in a multicriteria optimization context as well in this work.