A Modified LQG Algorithm (MLQG) for Robust Control of Nonlinear Multivariable Systems

Abstract

The original LQG algorithm is often characterized for its lack of robustness. This is because in the design of the estimator (Kalman filter) the process disturbance is assumed to be white noise. If the estimator is to give good estimates, the Kalman gain is increased which means that the estimator fails to become robust. A solution to this problem is to replace the proportional Kalman gain matrix by a dynamic PI algorithm and the proportional LQ feedback gain matrix by a PI algorithm. A tuning method is developed which facilitates the tuning of a modified LQG control system (MLQG) by only two tuning parameters