•
Introduction
Water systems are used for many purposes, i.e. navigation, agriculture, ecology, energy production, recreation.
Their daily management can be supported by controllable structures.
Model Predictive Control (MPC) is a valuable technique for the optimal management of such systems. MPC uses models to predict future effects of control actions in order to select the ones that will lead to the best predicted outcome over some limited horizon. Looking into the future makes the management anticipatory: the control actions proactively take in account expected future events and condition.
Example: Drainage Canal
Selection of optimal control according to the objective of maintaining the water level as close as possible to the target level along a future horizon
The Problem of the Uncertainty
Traditional MPC are optimized using one scenario of inflow, generally the average. Under extreme unexpected events the performance can be very poor.
The Problem of the Computation Time
If used in real time, the control action must be taken at periodic moment in time, hence the amount of time to solve the optimization problem is a limiting factor
Objective of the Research
Extending the capacity of MPC to have a control that is robust to uncertainty
•
available in a reasonable computation time •
Adaptive Control by use of Ensemble Forecast
Luciano Raso
1,2
– Dirk Schwanenberg
2
– Nick van Giessen
1
– Peter Jules Van Overloop
1
The control is optimal to the average inflow
Unsatisfactory performances under extreme events
At every moment of decision (every 15 min) the following procedure is adopted
Conclusion
The Control Tree has the advantage of being Adaptive: it recognizes presently the future possibility of changing the control according to the ongoing scenario. This postpones choices until more information is available and improves the control in the present.
Further Research:
optimal number of scenarios
A suitable ensemble reduction must produce scenario tree capable of representing well the entire set although a limited amount of branches
Contact Details
Section of Water Resources Management Delft University of Technology l.raso@tudelft.nl Complete Ensemble Scenario Tree Control TreeOptimal State Trajectories
1) Section of Water Resources Management
2) Dep. of Operational Water Management
Search for an optimal tradeoff
Ensemble
Generation
A prediction model produces a large number of equal probable scenarios that represent how uncertainty spreads in the future
Ensemble
Reduction
When scenarios (or parts of them) are sufficiently close to each other, they can be aggregated. This operation produces a Representative Scenario Tree, where every branch has its own probability
Topology
Inheritance
The Control Tree is created using the same topology of the Scenario TreeOptimal Control
The Control Tree is optimized respect to the Scenario Tree. The Optimal Control Tree provides the Optimal State Trajectories.
Implementation
of the Decision
The control action at the first time step is applied and the procedure is repeated
