Data-driven control-oriented reduced order modeling for open channel flows

Abstract

Model order reduction can be used for efficient simulation of complex systems. Data-based system identification approaches using neuronal networks or Dynamic Mode Decomposition enable us to extract characteristic properties of the system dynamics in order to reduce them to a low-dimensional space. There the temporal propagation can be described with significantly less computational effort. Both approaches are applied to the boundary actuated St. Venant equations, to obtain control-oriented reduced order models, which try to capture the dynamics of open channel flows for a wide range of input signals. It is investigated whether these models can be used for efficient simulation and how accurately they reconstruct the dynamic behavior of the water depth and velocity of an open channel.