Convex Optimization of Real-Time Reservoir Operation Considering Forecast Uncertainty

Optimization of water resources systems is complex due to many controlled variables, nonlinear and nonconvex structure of the problem and inherent uncertainties. In general, consideration of meteorological forecast uncertainty in short term decision support system is accomplished by taking multiple probabilistic streamflow forecasts (PSF) into decision support system. However, robust stochastic control approaches and their applications are still rare. Therefore, it is aimed to implement real time stochastic operation of a single multi-task reservoir in this study. The test reservoir has a gated spillway, limited reservoir volume and restricted downstream channel capacity. The two main tasks i.e. city water supply and flood control should be delivered. Tree based Model Predictive Control (MPC), the anticipatory control of the system and includes forecast uncertainty, is implemented with PSF scenarios under RTC-Tools 2 model. RTC-Tools 2 provides convex optimization using the Modelica language and goal programming. Convex optimization problem is set by convex objective function and convex feasible set, therefore it ensures global optimum and gives stable and deterministic results compared to non-convex solutions. However, this type of implementation requires dealing with nonlinearities such as volume-level relationship, gated spillway etc. The convex optimization of short-term stochastic MPC for a single event results are analyzed and compared with previously works done in the same region.