Sequential Model Predictive Control for Short-term Reservoir Optimization under Consideration of Hydraulic Constraints and Uncertain Inputs

We present a novel open source framework for implementing Sequential Model Predictive Control (MPC) for short-term reservoir optimization of low-lying polder systems. In the modular approach, the sequential execution of optimization and model simulation enables the integration of arbitrary sub-models with explicit and implicit time integrators. If the management of a water system is constraint by hydraulic conditions, i.e. limitations of pump capacities due to the flow capacity in the upstream canal network, in particular unconditionally stable implicit integrators appear useful in application to simplified hydraulic models. Our preference is the diffusive wave model, which is the simplest model for capturing backwater effects or the reversal of flow direction in the optimization approach. Another aspect of our work is the consideration of uncertain inputs, primarily provided on the basis of ensemble forecasts. The deterministic control trajectory of the MPC is therefore replaced by a scenario tree extending the approach to a multi-stage stochastic optimization. We focus on technical aspects of the implementation such as parallelization properties, since one of the major challenges is the runtime requirement of the extended approach in operational decision-support systems. The presentation of use cases in the Netherlands and elsewhere leads to a discussion on the skill of available ensemble forecasting products in application to the stochastic optimization procedure and its added value in comparison to the deterministic approach.