This dissertation proposes a Limited-Communication Distributed Model Predictive Control algorithm for networks with constrained discrete-time linear processes as local subsystems. The introduced algorithm has an iterative and cooperative framework with neighbor-to-neighbor communication structure. Convergence to a centralized solution is guaranteed by requiring coupled subsystems with local information to cooperate only. During an iteration, a local controller exchanges its predicted effects with local neighbors (which are treated as measured input disturbances in local dynamics) and receives the neighbor sensitivities for these effects at next iteration. Then the controller minimizes a local cost function that counts for the future effects to neighbors weighted by the received sensitivity information. Distributed observers are employed to estimate local states through local input-output signals. Closed-loop stability is proved for sufficiently long horizons. To reduce the computational loads associated with large horizons, local decisions are parametrized by Laguerre functions. A local agent can also reduce the communication burden by parametrizing the communicated data with Laguerre sequences. So far, convergence and closed-loop stability of the algorithm are proven under the assumptions of accessing all subsystem dynamics and cost functions information by a centralized monitor and sufficient number of iterations per sampling. However, these are not mild assumptions for many applications. To design a local convergence condition or a global condition that requires less information, tools from dissipativity theory are used. Although they are conservative conditions, the algorithm convergence can now be ensured either by requiring a distributed subsystem to show dissipativity in the local information dynamic inputs-outputs with gain less than unity or solving a global dissipative inequality with subsystem dissipativity gains and network topology only. Free variables are added to the local problems with the object of having freedom to design such convergence conditions. However, these new variables will result into a suboptimal algorithm that affects the proposed closed-loop stability. To ensure local MPC stability, therefore, a distributed synthesis, which considers the system interactions, of stabilizing terminal costs is introduced. Finally, to illustrate the aspects of the algorithm, coupled tank process and building HVAC system are used as application examples.
