A data driven approach using Takagi-Sugeno models for computationally efficient lumped floodplain modelling

Abstract:

Many applications in support of water management decisions require hydrodynamic models with limited calculation time, including real time control of river flooding, uncertainty and sensitivity analyses by Monte-Carlo simulations, and long term simulations in support of the statistical analysis of the model simulation results (e.g. flood frequency analysis). Several computationally efficient hydrodynamic models exist, but little attention is given to the modelling of floodplains. This paper presents a methodology that can emulate output from a full hydrodynamic model by pbkp_redicting one or several levels in a floodplain, together with the flow rate between river and floodplain. The overtopping of the embankment is modelled as an overflow at a weir. Adaptive neuro fuzzy inference systems (ANFIS) are exploited to cope with the varying factors affecting the flow. Different input sets and identification methods are considered in model construction. Because of the dual use of simplified physically based equations and data-driven techniques, the ANFIS consist of very few rules with a low number of input variables. A second calculation scheme can be followed for exceptionally large floods. The obtained nominal emulation model was tested for four floodplains along the river Dender in Belgium. Results show that the obtained models are accurate with low computational cost. © 2013 Elsevier B.V.

Año de publicación:

2013

Keywords:

• Computationally efficient
• Lumped modelling
• Floodplains
• Floodplain hydraulics
• Takagi-Sugeno
• Adaptive neuro fuzzy inference systems

Fuente:

scopus