Non-iterative distributed MPC for large-scale irrigation channels

This paper develops a non-iterative, distributed model predictive control (MPC) algorithm suitable for managing water-level references for large-scale automated irrigation channels to ensure the satisfaction of water-level and flow constraints. This algorithm is developed by first introducing the concept of spare supply capacity, which allows constraint information to be shared between serially-connected subsystems. Exact and approximate methods for calculating spare supply capacity are presented. A solution of the finite-horizon optimal control problem incorporating the computed spare supply capacity is then obtained for every subsystem. The collection of solutions obtained is shown to be “mutually feasible” in the sense that they describe a suboptimal solution of the large-scale problem. The computational complexity of the resulting distributed MPC algorithm is approximately linear in the number of control gates, making it potentially more efficient than a centralized MPC algorithm for long irrigation channels. A simulation investigation of part of an automated irrigation channel is presented which compares the proposed distributed MPC algorithm to existing MPC methods. This simulation investigation finds that while the proposed algorithm is successful in satisfying the imposed constraints, it yields higher peak gate flow commands than a centralized MPC controller. This is consistent with the trade-off of computational complexity for optimality exploited in moving from MPC to distributed MPC