Controller design and distributed control within multi-agent systems

An automatic process control framework based on MADCABS is developed and tested via simulations on various distributed control problems. MADCABS is an agent-based process supervision and control platform in Java. This research focused on control system design and distributed control with agent-based systems by using MADCABS.;Several controllers and control-related modules are developed in MADCABS as part of this research. The controllers include PID controller, model predictive control (MPC) systems, distributed MPC controllers, and distributed multi-rate controller. Other related modules include controller performance evaluation for single-input-single-output controllers and for MPC controllers, Kalman filters, and subspace system identification modules. The scheduling and communication between different modules are also designed and implemented.;A communication scheme for distributed MPC is developed. It relies on a coordinator in charge of communications between local distributed controllers and provides a flexible way to construct the local optimization problem in local controllers. An innovative method is also proposed to calculate the new setpoints for the target tracking problem under this coordination structure, without the need for a feasible starting point.;Modeling of multi-rate sampled systems is challenging yet critical for developing MPC systems. Traditional subspace space identification does not provide models for such systems with the desired accuracy for prediction. Under the coordination framework for distributed control in MADCABS, distributed system identification and control methods are proposed to control multi-rate sampled systems. The performance of the proposed multi-rate distributed control is tested the Tennessee Eastman challenge problem and the improvement of control is illustrated for various.