Exploring the Potential and the Boundaries of the Rolling Horizon Technique for the Management of Reservoir Systems with over-Year Behaviour

The paper presents a set of experiments on the Rolling Horizon Technique (RHT), applied to the management of a multi-reservoir, multipurpose water resources system with over-year behaviour. In the RHT, decisions on releases from reservoirs are taken for a number of time steps ahead (the Forecasting Horizon – FH) through an optimization model, based on the present water availability in reservoirs and on some forecasts of future inflows over the FH. Only the decision concerning the first time step (the current month) is then implemented, as new information on reservoir levels and forecasts becomes available so that the process can start again with updated information. The paper investigates how the quality of forecasts and the length of the FH impact on the effectiveness of the decisions taken. The evaluation exercise is carried out in two different demand scenarios considering some deficit-related indices, such as TSSD, the Sum of Squared Standardized Deficits, over an Operation Horizon (OH) of forty years. The scenarios are designed to evaluate the role of different demand levels (corresponding to higher/lower stress) on system’s behaviour. Results show that in systems with over-year behaviour, effective forecasts (here simulated through the best possible: actually occurred inflows from the current month up to the end of water year) seem to add little value to system’s operation compared to the most naive forecast (average monthly values), especially in contexts characterized by high pressure on resources. Where there is less pressure on resources, good-quality forecasts are more effective to improve operation. A FH of 24 months seems to enhance significantly system’s performances compared to a FH of 12 months, while FHs longer than two years produce no improvements in system’s performances. The paper also tries to define the limits of a finite-horizon operation technique such as the RHT by benchmarking it with an idealistic release schedule obtained by a model that performs allocation with Perfect Foresight over the whole Operation Horizon.