The short-term, optimal management of storage reservoirs is challenging due to multiple objectives, i.e. hydropower, water supply or flood mitigation, and inherent uncertainties of forecasts for inflow and water demand. Model Predictive Control (MPC) provides an online solution for this management problem by combining a process model, forecasts and the formulation of objectives in an objective function and its solution by an optimization algorithm. This anticipatory management has many advantages, but may suffer from forecast uncertainty. In practice, there are several sources of forecast uncertainty, which can jeopardize control decisions. In this study, hindcast experiments integrating deterministic and probabilistic streamflows in a closed-loop mode of MPC are tested to mimic a real-time flood mitigation case. Probabilistic inflow forecasts in combination with multi-stage stochastic optimization model are used with tree-based reduction techniques. According to the results, tree-based MPC proposes less spillway discharges during a real-time control of a major flood case by incorporating longer the forecast horizon and consideration of forecast uncertainty in the decision process. On the other hand, energy generation is compared with deterministic method, and the results are promising to be used without compromising the energy production.
Short-Term Reservoir Optimization for Flood Mitigation under Meteorological and Hydrological Forecast Uncertainty
