Multiple model iterative learning control

Abstract

Iterative learning controller (ILC), which is based on the model of the system, can give good performance in the steady state if the input is updated from trial to trial. The model, on which the ILC is set up, is not always needed to be exact. But a better model will lead to the deduction of iteration times. The iterative learning control problem of nonlinear system which can be described by a time invariant model and a variant model is studied. The model of time invariant part, which is known, is always used to set up ILC. The model of time variant part, which is always unknown, will be regarded as error model. Multiple fuzzy neural network (FNN) models are used to approximate the possible description of error model. By adding the multiple FNN models to the model of time invariant part, the model set of the system can be obtained. Every sample time, the best model will be decided by an index function which is formed by the output error between the model and the system. The ILC based on this model will improve the transient response of system greatly. The simulation will show the effectiveness of the method proposed in this paper.