Abstract In this study, an adaptive neural network control approach is proposed to achieve accurate and robust control of nonlinear systems with unknown dynamics, wherein the neural network is innovatively used to learn the inverse problem of system dynamics with guaranteed convergence. This study focuses on the following three contributions. First, the considered system is transformed into a multi-integrator system using an input-output linearization technique, and an extended state observation technique is used to identify the transformed states. Second, an iterative control learning algorithm is proposed to achieve the neural network training, and stability analysis is given to prove that the network’s predictions converge to ideal control inputs with guaranteed convergence. Third, an adaptive neural network controller is developed by combining the trained network and a proportional-integral controller, and the long-standing challenge of model-based methods for control determination of unknown dynamics is resolved. Simulation results of a virtual control mission and an aerospace altitude tracking mission are provided to substantiate the effectiveness of the proposed techniques and illustrate the adaptability and robustness of the proposed controller.

中文翻译：

---

动态未知非线性系统的自适应神经网络控制

摘要 在这项研究中，提出了一种自适应神经网络控制方法，以实现对未知动力学非线性系统的精确和鲁棒控制，其中创新地使用神经网络来学习具有保证收敛性的系统动力学逆问题。本研究侧重于以下三个贡献。首先，使用输入-输出线性化技术将所考虑的系统转换为多积分器系统，并使用扩展状态观察技术来识别转换状态。其次，提出了一种迭代控制学习算法来实现神经网络的训练，并通过稳定性分析证明网络的预测收敛到有保证收敛的理想控制输入。第三，通过结合训练网络和比例积分控制器开发了自适应神经网络控制器，解决了基于模型的未知动力学控制确定方法的长期挑战。提供了虚拟控制任务和航空航天高度跟踪任务的仿真结果，以证实所提出技术的有效性，并说明所提出控制器的适应性和鲁棒性。