This study considers the main challenges of presenting an iterative observer under a data-driven framework for nonlinear nonaffine multi-agent systems (MASs) that can estimate nonrepetitive uncertainties of initial states and disturbances by using the information from previous iterations. Consequently, an observer-based iterative learning control is proposed for the accurate consensus tracking. First, the dynamic effect of nonrepetitive initial states is transformed as a total disturbance of the linear data model which is developed to describe I/O iteration-dynamic relationship of nonlinear nonaffine MASs. Second, the measurement noises are considered as the main uncertainty of system output. Then, we present an iterative disturbance observer to estimate the total uncertainty caused by the nonrepetitive initial shifts and measurement noises together. Next, we further propose an observer-based switching iterative learning control (OBSILC) using the iterative disturbance observer to compensate the total uncertainty and an iterative parameter estimator to estimate unknown gradient parameters. The proposed OBSILC consists of two learning control algorithms and the only difference between the two is that an iteration-decrement factor is introduced in one of them to further reduce the effect of the total uncertainty. These two algorithms are switched to each other according to a preset error threshold. Theoretical results are demonstrated by the simulation study. The proposed OBSILC can reduce the influence of nonrepetitive initial values and measurement noises in the iterative learning control for MASs by only using I/O data.

本研究考虑了在非线性非仿射多智能体系统 (MAS) 的数据驱动框架下呈现迭代观测器的主要挑战，该系统可以通过使用来自先前迭代的信息来估计初始状态和扰动的非重复不确定性。因此，提出了一种基于观察者的迭代学习控制，用于准确的共识跟踪。首先，非重复初始状态的动态效应被转化为线性数据模型的总扰动，该模型被开发用于描述非线性非仿射 MAS 的 I/O 迭代-动态关系。其次，测量噪声被认为是系统输出的主要不确定性。然后，我们提出了一个迭代干扰观测器来估计由非重复初始偏移和测量噪声一起引起的总不确定性。接下来，我们进一步提出了一种基于观测器的切换迭代学习控制（OBSILC），使用迭代干扰观测器来补偿总不确定性，并使用迭代参数估计器来估计未知梯度参数。所提出的 OBSILC 由两种学习控制算法组成，两者之间的唯一区别是在其中之一中引入了迭代递减因子，以进一步降低总不确定性的影响。这两种算法根据预设的误差阈值相互切换。仿真研究证明了理论结果。