By considering non-repetitive uncertainties, an observer-based data-driven iterative learning control (ObDDILC) is proposed in this article for non-linear non-affine systems in the existence of non-repeatable disturbances, random initial values, and input constraints. Aiming to addressing the non-affine and non-linear characteristics of the systems, a linear data model (LDM) is constructed in iteration domain without introducing any physical interpretation but only for the purpose of the subsequent algorithm design and analysis. The iteration-varying initial values and disturbances are incorporated as a total non-repetitive uncertainty of the LDM. Both an iterative learning observer and a parameter estimator are proposed to address the total non-repetitive uncertainties and unknown parameters of the established LDM, respectively. Then, an observer-based learning control law is developed using the estimated output to compensate the impact of the non-repetitive uncertainties on the control performance, where a saturated function is employed to deal with the input constraints. The convergence of proposed ObDDILC is proved by using contraction mapping as the basic tool. All of the algorithms are designed and analysed without dependence of any model information except I/O data. The theoretical results are tested by simulations.

中文翻译：

---

基于观察者的数据驱动迭代学习控制

通过考虑非重复不确定性，本文针对存在不可重复扰动、随机初始值和输入约束的非线性非线性仿射系统，提出了一种基于观测器的数据驱动迭代学习控制（ObDDILC）。针对系统的非仿射和非线性特性，在迭代域构建线性数据模型（LDM），不引入任何物理解释，仅用于后续算法设计和分析。迭代变化的初始值和干扰被合并为 LDM 的总非重复不确定性。迭代学习观察器和参数估计器都被提出来分别解决已建立的 LDM 的总非重复不确定性和未知参数。然后，使用估计输出来开发基于观察者的学习控制律，以补偿非重复不确定性对控制性能的影响，其中采用饱和函数来处理输入约束。通过使用收缩映射作为基本工具证明了所提出的 ObDDILC 的收敛性。所有算法的设计和分析不依赖于除 I/O 数据之外的任何模型信息。通过仿真对理论结果进行了测试。所有算法的设计和分析不依赖于除 I/O 数据之外的任何模型信息。通过仿真对理论结果进行了测试。所有算法的设计和分析不依赖于除 I/O 数据之外的任何模型信息。通过仿真对理论结果进行了测试。