Data Assimilation to Improve Models used for the Automatic Control of Rivers or Canals

The dams and the hydropower plants on the Rhone River, managed by the Companie Nationale du Rhone (CNR), are controller in real-time by Model Predictive Controllers (MPC) since the early 2000s. The control objectives and constraints are manyfold: optimize electrical production, allow navigation, protect the banks from erosion, prevent or reduce the damages during flood events, supply water to industries, cities and irrigation districts. In case the outputs of the embedded model used by MPC do not fit the field measurements, some questions are raised on: how to interpret this, and what can be done to solve this problem? We will present recent developments, carried out and illustrated on the Rhone River allowing to address these issues. The framework we will use is the one of Kalman filtering. We will see that this framework is very powerful to solve the above described problems. But, in some cases the obtained solution is not the one we would expect. The conditions of success can be expressed and checked from some mathematical tests, and linked to some physical properties (number and location of sensors, uncertainties of the measurements and of the model, hydraulic configuration of the hydraulic system).