MIMO Discrete Time Feedback Error Learning by Means of the Desired Value of the Future

Abstract

This paper proposes an improved learning scheme for DTFEL (Discrete time feedback error learning) . Although conventional DTFEL schemes based on the gradient method are easily implemented, these schemes are applied to only SISO systems and have drawbacks that assumptions of positive realness and PE condition for the convergence of the tracking error are required. Also, a class of feedforward (FF) controller is confined to biproper case, and the prefilter parameters, which is known, have to be learned as well as the plant parameters. On the other hand, the proposed scheme can relax two assumptions above, be applied to MIMO systems and improve the problems of both degree-of-freedom and redundancy of FF controller by using the desired value of the future. The paper verifies its effectiveness in terms of simulation.