Experimental results for distributed model predictive control applied to a water distribution system

In this paper, experimental results are presented for a distributed model predictive control (DMPC) scheme applied to a laboratory-scale water distribution system consisting of connected water tanks. The setup is an example of coupled dynamical systems whose modular structure makes them candidates for agent based distributed control methods. A set of low cost Raspberry Pi microcomputers connected via a common Ethernet-network is used for the implementation of the DMPC. The DMPC algorithm employs an augmented Lagrangian approach to solve the distributed optimal control problem (OCP) subject to nonlinear system dynamics in continuous-time form and input constraints. In this way, the performance and modular character of the DMPC as well as the individual runtime footprints of the communication and computation with respect to the overall runtime of the distributed algorithm are studied.