Plantwide cooperative distributed model predictive control

This thesis presents a plantwide model predictive control strategy in which subsystems provide feedback by solving a local optimization and exchanging information across the plant. The plant model is required to satisfy a stabilizability condition that does not depend upon the strength of the open-loop, inter-subsystem interactions. This control strategy has the following features: hard input constraints are satisfied; terminating the distributed optimization iteration prior to convergence does not affect nominal stability; the distributed optimization converges to the Pareto optimal (centralized) solution; no coordinating optimization is employed. Exponential stability of the closed-loop plantwide system is proven for the state and output feedback cases. A modification is presented in which constraints sparsely coupled between subsystems can be handled without loss of stability or optimality.;An extension to the cooperative controller is provided in which communication between subsystems occurs at multiple time schedules, as in traditional hierarchical control. The hierarchical control scheme is shown to provide exponentially stable performance under state and output feedback. This extension uses a modification of the distributed optimization. The optimization is shown to converge to the Pareto optimum and can be terminated early without affecting nominal stability.;To demonstrate the flexibility of cooperative control, a series of examples are presented. These examples show that cooperative control can nominally stabilize subsystems optimizing, communicating, and sampling on any time scale. Further examples are provided to illustrate that the modeling required for cooperative control can be identified over time and that the models can be generated from a centralized first-principles model.;Finally, a novel distributed optimization for nonconvex problems is presented that provides convergence to stationary points. This optimization is combined with a new result in suboptimal MPC to develop a plantwide distributed nonlinear controller. This controller provides asymptotically stablizing feedback and an example is used to illustrate performance.