A Tree-Based Model Predictive Control Implementation on the North Sea Canal—Amsterdam-Rhine Canal

We here report an interesting case study of an evaluation of a tree-based model predictive control solution (TB-MPC) to advice operators on the optimal response to a forecast compound flood event. Tree-based model predictive control was implemented in the decision support system for the North Sea Canal, the Netherlands and tested numerically using a 1D hydrodynamic model and measured data. The North Sea Canal is connected to the North Sea by the largest pumping station of the Netherlands at IJmuiden and has uncertain inflows coming from ensemble weather forecasts applied with a rainfall-runoff model. While the canal conducts the drained water from the lowlands to the sea, the water levels should be kept within a certain range for safety and shipping, and the cost of pumping should be as low as possible. Deterministic (single forecast) linear MPC is currently used in operations to determine when to use each individual pump of the pumping station and when to spill water under gravity through the sluices next to the pumping station. Now that ensemble inflow forecast has become available for this system, we set to explore the feasibility, advantages, and limitations of using TB-MPC with a linear optimization model adapted from the original model. This case-study is interesting because: (1) Real data is used on a real decision support system to perform a closed-loop analysis for two flood events using tree-based as well as deterministic MPC. (2) Applied case study for a real rainfall event. (3) It is integrated into a decision support system: in case it is desired, just by pushing a button the pump settings computed by tree-based MPC are automatically executed.