November 8, 2022
@
4:00 PM
–
5:00 PM
Alexander Mitsos
Professor and Laboratory Director
Process Systems Engineering
Aachener Verfahrenstechnik (RWTH Aachen University)
Aachen, North Rhine-Westphalia
Germany
Global Optimization with Embedded Hybrid Models
We present on theory, algorithms and applications of reduced-space formulations for the deterministic global optimization with hybrid mechanistic/data-driven models embedded.
Within the broader scope of digitalization of the (bio)chemical industry, surrogate models are gaining increasing attention. We first discuss the need for hybrid modeling combining the best of both mechanistic and data-driven models. We present our work in this direction both from the software and algorithmic development side. We discuss our “Machine Learning Models for Optimization (MeLOn)” toolbox that enables integration of data-driven models to optimization problems.
These problems are then solved in reduced space by our deterministic global optimization software “McCormick-based Algorithm for mixed-integer Nonlinear Global Optimization (MAiNGO)”. We discuss in detail artificial neural networks, gaussian processes andHammerstein-Wiener models. We demonstrate the advantages of our proposed approach for problems from process systems engineering, in particular from flowsheet optimization.