Trajectory tracking control of discrete non-affine MIMO iterative systems with unknown models: a neural-network-based data-driven algorithm

This paper devises a neural-network-based data-driven (NN-DD) algorithm to address the trajectory tracking control (TC) of discrete non-affine MIMO systems with unknown models and repetitive operation patterns. Data-driven control no longer relies on the precise model of the controlled system, thereby breaking free from the limitations of model-based control strategies. Inspired by this, the primary objective of the algorithm is to ensure that the tracking error of the system is uniformly ultimately bounded through a data-driven approach. The algorithm is comprised of a DD modeling approach based on an enhanced stochastic configuration network (ESCN), and a control input solving approach based on radial basis function neural networks (RBFNNs). The numerical simulations indicate that the proposed algorithm achieves a decrease in the modeling error to \(3.29e-7\) $3.29e-7$ and the tracking error to \(1e-8\) $1e-8$ after just 20 iterations. In addition, the numerical simulations also demonstrate that the modeling algorithm based on an ESCN reduces the modeling errors by \(48.11\%\) $48.11\%$ and \(99.95\%\) $99.95\%$ respectively compared to the modeling algorithms using only stochastic configuration networks (SCNs) or extreme learning machines (ELMs). Regarding tracking errors, the proposed RBFNN-based controller reduces the tracking error by \(100\%\) $100\%$ compared to the backpropagation NN (BPNN)-based controller. Furthermore, the robustness of the algorithm against time-varying interference is tested via the unmanned vehicle simulations. This paper covers several contributions: 1) The proposed algorithm is entirely DD and can directly establish the relationship between inputs and outputs. 2) The designed ESCN fully integrates the advantages of SCNs and ELMs, in contrast to simply combining basic algorithms. 3) The RBFNN-based controller is independent of the actual system structure and exhibits excellent generalization capabilities.