[3H1-OS-9a] OS-9

9:00 AM - 9:15 AM JST(12:00 AM - 12:15 AM UTC)

[3H1-OS-9a-01] Control and Stability Evaluation of Complex Dynamical Systems Using Neural ODEs

〇Satoshi Ueda¹, Hideaki Ogawa¹ (1. Japan Aerospace Exploration Agency)

Keywords:

Neural ODE,Complex dynamical systems,Control,Stability

Optimal control for complex dynamical systems is a useful method for solving various problems in natural and applied sciences. While optimal control problems can be formulated using the variational principle, solving complex dynamical systems analytically is often difficult. Neural ordinary differential equations (ODEs) are a deep learning framework capable of representing continuous dynamics. Previous research has proposed frameworks that automatically learn control signals using neural ODEs, demonstrating the ability to transition dynamical systems to target states within specified time limits. This research proposes a control framework based on neural ODEs for complex dynamical systems, providing a solution to the optimal control problem with state feedback. It demonstrates the ability to robustly control systems from initial states with certain errors toward target terminal states. Furthermore, it presents a comprehensive framework for evaluating the stability of closed-loop systems incorporating the control law, utilizing a data-driven Lyapunov equation solution and contraction analysis.