Model Reference Learning Control for Nonlinear Systems

摘要

This paper proposes a model reference learning control for nonlinear plants. The nonlinearity is estimated through trials, which start at a same initial condition. The control input is generated on the basis of the estimates, so as to force the output of the plant to follow the output of the prespecified reference model which is driven by an arbitrary bounded command input. Unlike model reference adaptive control, this controller can cope with nonlinearity because the controller estimates optimal values of tunable functions, adding to parameters. Moreover, the command input can be varied through trials since nonlinear functions included in the model reference controller are adjusted through trials, whereas the repetitive learning controller cannot, since it estimates the control input directly. Convergence of the following error is analysed theoretically and is verified by a simple numerical simulation on a one-pivot manipulator.