Cyber-physical systems integrate computing, network and physical environments to make the systems more efficient and cooperative, and have important and extensive application prospects, such as the Internet of things. This paper studies the control problem of nonlinear cyber-physical systems with unknown dynamics and communication delays. A networked learning predictive control scheme is proposed for unknown nonlinear cyber-physical systems. This scheme recursively learns unknown system dynamics, actively compensates for communication delays and accurately tracks a desired reference. Learning multi-step predictors are presented to predict various step ahead outputs of the unknown nonlinear cyber-physical systems. The optimal design of controllers minimises a performance cost function which measures the tracking error predictions and control input increment predictions. The system analysis leads to the stability criteria of closed-loop nonlinear cyber-physical systems employing the networked learning predictive control scheme. An example illustrates the outcomes of the proposed scheme.

网络物理系统集成了计算，网络和物理环境，以使系统更加高效和协作，并具有重要而广泛的应用前景，例如物联网。本文研究了动力学和通信时延未知的非线性网络物理系统的控制问题。针对未知的非线性网络物理系统，提出了一种网络学习预测控制方案。该方案递归地学习未知的系统动力学，主动补偿通信延迟并准确跟踪所需的参考。提出了学习多步预测器，以预测未知非线性网络物理系统的各种提前输出。控制器的最佳设计可最大程度地降低性能成本函数，该函数可测量跟踪误差预测和控制输入增量预测。系统分析得出了采用网络学习预测控制方案的闭环非线性网络物理系统的稳定性标准。一个例子说明了该方案的结果。