A New Neural ODE Structure for Learning High-Order Dynamical Systems
A New Neural ODE Structure for Learning High-Order Dynamical Systems
复制标题
DOI：
10.1109/sose55472.2022.9812703
复制
发表时间：
2022-06
期刊：
影响因子：
--
通讯作者：
Shiqi Nan;C. Qian
中科院分区：
文献类型：
--
作者：
Shiqi Nan;C. Qian
动态系统是对世界上各种应用的数学描述。然而，动态系统的控制存在着非线性、不确定性和高维性等诸多挑战。最近的研究揭示了神经网络和动力系统之间的重要联系。神经网络是用于学习和预测动态系统的强大技术。相应地，动态洞察可以应用于神经网络。本文研究了神经网络结构对高阶动力系统的学习。提出了一种基于神经常微分方程的连续高阶神经网络结构，用于高阶平面动力系统的建模。
Dynamical systems are mathematical descriptions of applications around our world. However, there are many challenges in control of dynamical systems, such as nonlinearity, uncertainty and high dimensionality. Recent research has revealed significant connections between neural networks and dynamical systems. Neural networks are powerful technologies that used for learning and predicting dynamical systems. Correspondingly, dynamical insights could be applied to neural networks. In this paper, we investigated neural network structures to learn highorder dynamical systems. We proposed a continuous high-order neural network structure based on Neural Ordinary Differential Equations to model high-order planar dynamical systems.