Presentation of two optimal controllers

Abstract

A comparison is made between a continuous-time generalized predictive controller (CGPC) and a linear quadratic Gaussian (LQG) controller. The nonzero set-point LQG controller is designed under the assumption that the system to be controlled is modeled without error and that any system disturbances are white random processes. However, the basic LQG regulator may be inadequate when these conditions are violated. Control system performance then may be improved by adding new states to the closed-loop system (i.e., by providing dynamic compensation). The nonzero set-point LQG controller can be combined with the continuous-time identification procedure in adaptive control schemes. Simulation results illustrate the proposed controller design procedure.<>