Model-free Iterative Learning Control based on Adaptive Dynamic Programming

An exploration of the model-free iterative learning control based on adaptive dynamic programming is developed for nonlinear non-affine discrete-time systems. A data-driven iterative dynamic linearization model is developed for original nonlinear systems only using input/output data. Subsequently, a novel performance index function incorporating the error information of future batches is formulated along the iteration axis through dynamic programming, where neural networks are employed to estimate future information. The gradient descent algorithm is employed to facilitate the weight updating of the neural network. Ultimately, a variable penalty factor is developed. This synthesis collectively enhances the convergence rate of tracking errors. The effectiveness of this method is substantiated through two simulation examples.