Data-Driven Predictive Iterative Learning Control for Nonlinear Non-Affine Systems With Unknown Fading Channels

Abstract

With the rapid advancement of network technology, networked control systems (NCSs) have found extensive applications in recent years. However, fading channels are widespread in NCSs, which can greatly impair control performance. In this paper, the data-driven predictive iterative learning control (DDPILC) scheme with fading estimation and compensation algorithms for nonlinear non-affine NCSs is investigated to address the control issue under the influence of fading channels on both input and output sides of the systems. Firstly, this paper proposes iterative estimators based on stochastic gradient. The proposed estimators can estimate the statistical information of fading channels so that their adverse effect can be compensated for by designing of fading compensation algorithms. Secondly, the control law and PPD estimation and prediction algorithms are designed, forming the DDPILC scheme. The designed DDPILC scheme can not only leverage the historical input and output data of the systems but also predict the future pseudo partial derivative (PPD), fusing the dynamic adjustment capability of predictive control and with the high tracking accuracy of ILC for repetitive systems. Thirdly, the convergence analysis shows that the system’s output converges along the iteration axis with high precision under the influence of fading channels on both sides. Finally, a simulation of permanent magnet linear motor (PMLM) and a practical experiment on a networked liquid level control system validate the efficacy of the proposed DDPILC. Note to Practitioners—In recent years, Industrial Internet has developed rapidly. Against this background, networked control systems (NCSs) have been widely applied. However, fading channels in data transmission significantly degrade the tracking performance of NCSs. To address this issue, a data-driven predictive iterative learning control (DDPILC) scheme integrated with fading estimation and compensation algorithms on both input and output sides are proposed in this paper. This scheme requires no prior knowledge of fading channels’ statistics and systems, ensuring that the output of NCSs tracks the desired trajectory with high precision under fading channels on both sides. Simulation on a permanent magnet linear motor (PMLM) and experiment on a liquid level control system are provided to verify the effectiveness of the proposed scheme in practical applications. Therefore, this paper establishes a reliable basis for the practical implementation of the DDPILC strategy in NCSs with fading channels.