PID Controllers as Data Assimilation Tool for 1D Hydrodynamic Models of Different Complexity

Abstract

Flood risks management is based on data obtained by different forecasts. Forecasts are often based on hydrological-hydrodynamic models. These models are calibrated using selected time-series from the past. However, even calibrated models in later exploitation phases can produce solutions of unsatisfying accuracy. Some of the reasons are uncertainty in the initial and boundary conditions, uncertainty of the input data and uncertainty in riverbed geometry. The aim of the assimilation is to improve the results obtained from the previously calibrated model by coupling it with observed data. To assimilate, the model is run for a short previous period and the state of the model is adjusted to observed data. The corrected model state is then used as an initial state to run the model with for short-term forecast of input data. Assimilation method based on the PID controller for 1D river hydrodynamic models is analyzed in this paper. This method adjusts the state in the hydrodynamic models according to the measurements indirectly by adding or subtracting the discharge in the junction/sections with measured water level. The influence of the hydrodynamic model complexity is analyzed, comparing three models: non-inertia model, diffusion wave and dynamic wave model. Results show that PID control can be adequately used even coupled with simplified hydraulic models for short-term assimilation and forecast, without significant loss of accuracy. PID control-based data assimilation also yields significant reduction in the computational runtime.