On a self-tuning controller with retained and changeable memory

To enhance the robustness of self-tuning control, a classifier based on correlation is suggested as the underlying decision mechanism of a coarse estimator. Using the proposed state-class and parameter-class matrices as memory elements, both a priori and online measured (or estimated) information are retained and can be updated according to changing experimental conditions. This general framework of the coarse estimator is integrated with a previously developed fine estimator structure to provide reliable parameter estimates for a LQ-controller based on receding-horizon cost functions. The modified self-tuning controller has been found by simulation to give much improved results.