Model induction from data: towards the next generation of computational engines in hydraulics and hydrology; Proefschrift Technische Universiteit Delft.

There has been an explosive growth of methods in recent years for learning(or estimatingdependency) from data, where data refers to the known samples that are combinations of inputs and corresponding outputs of a given physical system. The main subject that is addressed in this thesisis, therefore, model induction from data for the simulation of hydrodynamic processes in the aquatic environment. First, some currently popular artificial neural network architectures are introduced, and it is then argued that these devices can be regarded as domain knowledge in capsulators by applying the method to the generation of wave equations fro m hydraulic data and showing how the equations of numerical-hydraulic models can, in their turn, be recaptured using artificial neural networks.The thesis also demonstrates how artificial neural networks can be used to generate numerical operators on non-structured grids for the simulation of hydrodynamic processes in two-dimensional flow systems and a methodology has been derived for developing generic hydrodynamic models using artificial neural network.The thesis also highlights one other model induction technique, namely that of the support vector machine, as an emerging new method with a potential to provide more robust models