Limited-Communication Distributed Model Predictive Control for Coupled and Constrained Subsystems

Abstract

This brief presents the limited-communication distributed model predictive control (LC-DMPC) algorithm for coupled and constrained linear discrete systems. In this framework, the global control problem is divided into a number of coupled subsystems based on a neighbor upstream and downstream structure. According to this structure, a subsystem views the coupling signals from upstream neighbors as measured disturbances and at the same time has outputs to downstream neighbors. In contrast with most DMPC schemes, the individual subsystems solve a different cost than the centralized problem. At each iteration, two bidirectional signals are communicated: predicted disturbances for downstream neighbors and local cost sensitivity to disturbances from upstream neighbors. With only neighbor-to-neighbor communication, the LC-DMPC scheme can converge to the centralized optimum without sharing dynamics or costs between the distributed agents. The closed-loop stability is guaranteed by assuming sufficiently long horizons. To reduce the computational burden, local control actions are parameterized by Laguerre functions. This leads to smaller distributed control problems and allows more iterations per sampling. A coupled tank process demonstrates the main aspects of the proposed algorithm.