DNN-based Implementation of Data-Driven Iterative Learning Control for Unknown System Dynamics

Abstract

As the condition of iterative learning control, it is usually necessary to estimate the parameters of the system model to determine whether the system satisfies the global Lipschitz condition and estimate the upper and lower bounds of the rate of change of the system. However, for systems with unknown dynamics, the data-driven iterative learning control based on system input and output cannot be realized fully. In this paper, using the nonlinear mapping and feature extraction ability of deep learning, only input / output data is used to determine whether the uncertain system satisfies the global Lipschitz condition and estimate the upper and lower bounds of the system's rate of change, so as to realize the iterative learning control of the system. The simulation results verify the validity of estimating whether the system satisfies the ILC condition only based on the input / output data of the system.