Keywords
real-time control, reservoir operation, anticipatory management
Start Date
1-7-2010 12:00 AM
Abstract
The use of Stochastic Dynamic Programming for designing optimal regulation policiesfor water reservoirs networks is well established in the literature. However, the calculation of theoptimal solution for many real problems is prohibitively time-consuming and this has motivatedapproximations, particulary in the description of the reservoirs inflow process. Furthermore thevariability of climatic forcing and the continuous adaptation in the management of water resourcesrequired by economic iusses represent a challenge for developing adapting real-time optimizazionprocedures. In this paper a real time control approach is proposed and applied to the case study ofthe multiprupose regulated lake Lugano, Switzerland. The lake operation problem is first solved atplanning level (off-line), using ciclostationary long-term statistics of the inflow, and is then refinedon-line using inflow forecast from a dynamical, non-linear, heteroscedastic model, that providesboth the expected value and the standard deviation of the inflow forecast. The model is forced byall the hydrometeorological information available in real time, both rainfall and runoff measuredat previous time steps. The effectiveness of the model in terms of the reservoir regulation is evaluatedthrough simulation and comparison with the results provided by conventional homoscedasticinflow models and with a ”perfect” forecast. Simulation results show that real-time control cansignificantely improve the system performances, especially for the purpose of flood control. Theimprovements are more significant by increasing the forecasting horizon.
Coupling hydrological models and wheather forecast for improved real-time management of water resources
The use of Stochastic Dynamic Programming for designing optimal regulation policiesfor water reservoirs networks is well established in the literature. However, the calculation of theoptimal solution for many real problems is prohibitively time-consuming and this has motivatedapproximations, particulary in the description of the reservoirs inflow process. Furthermore thevariability of climatic forcing and the continuous adaptation in the management of water resourcesrequired by economic iusses represent a challenge for developing adapting real-time optimizazionprocedures. In this paper a real time control approach is proposed and applied to the case study ofthe multiprupose regulated lake Lugano, Switzerland. The lake operation problem is first solved atplanning level (off-line), using ciclostationary long-term statistics of the inflow, and is then refinedon-line using inflow forecast from a dynamical, non-linear, heteroscedastic model, that providesboth the expected value and the standard deviation of the inflow forecast. The model is forced byall the hydrometeorological information available in real time, both rainfall and runoff measuredat previous time steps. The effectiveness of the model in terms of the reservoir regulation is evaluatedthrough simulation and comparison with the results provided by conventional homoscedasticinflow models and with a ”perfect” forecast. Simulation results show that real-time control cansignificantely improve the system performances, especially for the purpose of flood control. Theimprovements are more significant by increasing the forecasting horizon.
