Feedforward learning control for SISO plant with finite zeros and nonlinearity

Abstract

This paper proposes a scheme for feedforward (FF) learning control based on scheduled locally weighted regression (S-LWR). For an unknown nonlinear single-input single-output (SISO) plant, it generates FF control signals based on local models of inverse dynamics identified by on-line S-LWR learning at a current operating point (called scheduling parameter). This scheme was proposed previously by the authors under the assumption that every linear approximation of the plant is free of finite zeros; i.e., the numerator of its transfer function is constant. The objective of this paper is to relax this restriction by providing adjustable FF controller poles to cancel the plant zeros. Numerical simulation is carried out to verify effectiveness of the propose scheme.

Feedforward control with on-line tuning: A perspective on two-degree-of-freedom structure

“…In these works, the plant is assumed to be unknown but linear time-invariant. The learning law has been further generalized to nonlinear or time-variant plant via locally weighted regression with the linear filter [3], [9]. Experimental validation of linear filter FEL has also been reported; e.g., [5].…”

Feedforward learning control for MIMO plant with finite zeros: Parameterization of numerator polynomial matrix

“…Motivated by this model, called "feedback error learning (FEL)," much effort has been made to apply it to control system design in the past decade; see, e.g., [2], [3], [4], [5]. The learning law is further generalized to nonlinear plant via locally weighted regression [6], [7].…”