Real time optimal control of a reservoir system with probabilistic streamflow forecasts

Real time optimal control of reservoirs is a challenging task due to increasing water and energy demands, changing hydroclimatic conditions, complex system structure and uncertainties in the system. In this study, the problem is solved in several stages. In the first one, an operating technique is proposed by a joint approach using feedback and predictive controls. To that end, Implicit Stochastic Optimization is employed to derive a characteristic Guide Curve (GC) using historical data. Later, GC is tested and improved according to drought and flood inflow scenarios. Short term flood control is accomplished by Model Predictive Control (MPC) and the model performances are tested using perfect and perturbed forecasts via hindcast experiments. This part shows us, the control robustness can be affected by inconsistent forecasts during a real-time operation. At this point, second stage is intended for consideration of forecast uncertainty; thus Probabilistic Streamflow Forecasts (PSF) are generated by a new synthetic method and utilized in the optimization. An up-to-date technique is multi-stage stochastic MPC using scenario trees, referred to as Tree-Based MPC (TB-MPC). The methodology is applied to a test case which requires a challenging spillway operation due to the restricted downstream channel capacity. As a result, TB-MPC outperforms deterministic counterpart MPC in terms of minimizing downstream flooding risk according to maximum peak flow and flood volume indicators without compromising the water supply and energy generation. Finally, the study shows operational improvement of a reservoir system by probabilistic forecasts and latest optimal control methods.