Direct data-based model predictive control with applications to structures, robotic swarms, and aircraft.

摘要

A direct method to design data-based model predictive controllers is presented. The design method uses system identification techniques to identify model predictive controller gains directly from a set of excitation input and disturbance corrupted output. The design is direct in that the controller gains can be designed directly from input and disturbance corrupted output data without an intermediate identification step. The direct design is simpler than previous two-step designs and reduces computation time for the design of the controller. The direct design also enables an adaptive implementation capable of identifying controller gains online. The direct data-based controllers can be used for vibration suppression, disturbance rejection, tracking and is applied to structures, robot swarms and aircraft. For the cases of vibration suppression and disturbance rejection, the data-based controller has the advantage that any disturbances present in the design data are automatically rejected without needing to know the details of the disturbances. For the case of robot swarms, extensions are made for formation control and obstacle avoidance, and the controller can be implemented as a decentralized controller in real time and in parallel on individual vehicles with communication limited to past input and past output data. A formulation for improving the robustness of the controller to parametric variations is also developed. Finally, the adaptive implementation is shown to be useful for the control of linear time-varying systems and has been successfully implemented to control a linear time-varying model of a Cruise Efficient Short Take-Off and Landing (CESTOL) type aircraft.