Primal and dual decomposition for distributed MPC — Theory, implementation, and comparison in a SoS simulation framework

This paper presents two hierarchical techniques for the coordination of distributed model predictive controllers and discusses and compares their robustness to the presence of local constraints and their speed of convergence. These techniques follow feasible- and infeasible-path coordination approaches. The controlled system is assumed to consist of subsystems which are only connected by the use of shared resources. The goal is to achieve an optimal distribution of these resources so that an optimal operation of the overall system is achieved. The algorithms are applied to a benchmark problem of distributed model predictive control and the two coordination approaches are compared. The controllers and the model of the benchmark are implemented in a novel Modelica-based software framework for simulation-based validation of systems of systems that aims at reducing the engineering effort while systems of systems that aims at reducing the engineering effort while facilitating model re-usability and the deployment of distributed controllers in real-world industrial systems.