Coordinating multiple model predictive controllers for the management of large-scale water systems

The optimal management of multi-purpose water reservoir networks is a challenging control problem, because of the simultaneous presence of multiple objectives, the uncertainties associated with the inflow processes and the several interactions between the subsystems. For such systems, model predictive control (MPC) is an attractive control strategy that can be implemented in both centralized and decentralized configurations. The latter is easy to implement and is characterized by reduced computational requirements, but its performance is sub-optimum. However, individual decentralized controllers can be coordinated and driven towards the performance of a centralized configuration. Coordination can be achieved through the communication of information between the subsystems, and the modification of the local control problems to ensure cooperation between the controllers. In this work the applicability of coordination algorithms for the operation of water reservoir networks is evaluated. The performance of the algorithms is evaluated through numerical simulation experiments on a quadruple tank system and a two reservoir water network. The analysis also includes a numerical study of the trade-off between the algorithms' computational burden and the different levels of cooperation. The results show the potential of the proposed approach, which could provide a viable alternative to traditional control methods in real-world applications.