Limited-communication distributed model predictive control for output feedback coupled subsystems

This work introduces a distributed model predictive control algorithm for output feedback coupled and constrained subsystems. The proposed algorithm is based on neighbor communications between the directly coupled agents. Distributed Kalman filters are employed to estimate the local states through local outputs, control actions, and measured disturbances. This is an iterative approach where local agents cooperate with neighbors and convergence to the systemwide optimum depends only on this cooperation. The local controllers use only local information to solve the distributed optimization problems. The minimal required communication and the independent nature of local agents make the approach highly modular. To reduce the communication load, each agent in the network parametrizes the communicated vectors with Laguerre functions. This results in a reduced vector length and local agents share only Laguerre coefficients and a time constant. To illustrate the aspects of the algorithm, a coupled six-tank process is used as an application example.