Short term reservoirs operation on the Seine River: performance analysis of Tree-Based Model Predictive Control
Opération à court terme des réservoirs de la Seine: analyse des performances de la commande prédictive Tree-Based Model Predictive Control
Résumé
The Seine River, in France, flows through territories of large economic value, among which the metropolitan area of Paris. A system of four reservoirs operates upstream to regulate the river flows in order to protect the area against extreme events, such as floods and droughts. Current reservoirs management is based on reactive filling curves, designed from an analysis of historical hydrological regimes. The efficiency of this management strategy is jeopardized when inflows are significantly different from their seasonal average. To improve the current management strategy, we investigated the use of Tree-Based Model Predictive Control (TB-MPC). TB-MPC is a proactive and centralized method that uses information available in real-time, as ensemble weather forecasts. Reservoir management is tested under past hydro-climatic conditions using time series of ensemble weather forecasts produced by ECMWF (European Centre for Medium-Range Weather Forecasts) and weather observations. The performance of TB-MPC is compared to that of deterministic Model Predictive Control (MPC), showing the benefits of considering forecasts uncertainty by using ensemble forecasts.