Quasi-decentralized output feedback model predictive control of networked process systems with forecast-triggered communication

This work presents a framework for quasi-decentralized output feedback model predictive control (MPC) design with an adaptive forecast-triggered communication strategy. Based on distributed Lyapunov-based control, an MPC controller is initially designed for local control of every subsystem in the entire networked process system. A supervisory observer that has access to the process input and output information generates estimates of the process state. And the state estimates can be used to update the model states of the local controllers and we show that this quasi-decentralized MPC design is able to practically stabilize the entire networked process system if the model states are updated at every sampling instant. In order to minimize the communication from the supervisory observer to the local control systems, an adaptive forecast-triggered communication strategy is proposed. A key idea of this strategy is to forecast the future evolution of each subsystem and generate a worst-case estimate by using the closed-loop stability properties as well as the information about the current operating status of each subsystem. Whenever the forecast indicates possible instability in the future, the observer estimate will be immediately transmitted to update the model state within the control system that needs attention in order to preserve stability; if the forecast shows no signs of instability, then the local control system will continue to rely on the model. The implementation of the developed methodology is demonstrated using a simulated model of a chemical process.