Doctoral dissertation of Ward Quaghebeur – Hybrid models of dynamical systems: neural differential equations, Shapley value analysis and illustrations in water systems

Ward Quaghebeur happily posing next to the jury members

On 11 January 2023, Ward Quaghebeur successfully defended his Doctoral Thesis, another milestone in BIOMATH.

About Ward:

After having obtained his master degree in Bioscience Engineering with highest honours at Ghent University, Ward started his research career at KU Leuven, where he worked shortly on purification methods for drinking water. He then combined a PhD program at Ghent University with a second master degree in Computer Science at the Georgian Institute of Technology. He has been the first author of a number of peer-reviewed articles and co-authored several other publications, both at renowned scientific journals and international conferences. Moreover, he has recently participated in several water-related seminars and workshops and obtained further recognition, particularly in the form of publication awards and scientific grants.

Summary of the Doctoral Thesis:

Hybrid modelling, combining mechanistic and data-driven components, is proposed as a way to address the shortcomings of each modelling paradigm and enhance the design, comprehension, and operation of water systems. This new framework incorporates a neural network into the differential equations of the mechanistic model and has been tested on various systems, showing its ability to extrapolate and fill gaps in domain knowledge. Shapley value analysis is introduced to analyse the shortcomings of data-driven models and identify areas of improvement. It is demonstrated on a variety of systems, including an Activated Sludge Model and two real-world datasets, to successfully retrieve missing dynamics and highlight the deficiencies of mechanistic models.

Internal contact(s)