Physics-informed neural ODE for periodic orbits continuation

Authors:

Abstract
To overcome the vulnerability of traditional numerical continuation methods when computing periodic orbits with extreme initial-value sensitivity in the Circular Restricted Three-Body Problem (CR3BP), this paper proposes a Physics-Informed Neural ODE (PINO) framework for orbit family continuation. The method uses Neural ODEs as a learnable dynamics solver and embeds periodic boundary conditions and orbital symmetry constraints into the loss function to enforce physical consistency during orbit correction. A parameter mapping network takes initial position as input and outputs period and initial velocity, enabling continuous prediction of orbit parameters after physics-enhanced training. The iterative “predict–refine–expand” workflow progressively covers the target interval while ensuring accuracy at each point. Simulation results confirm that the PINO framework effectively achieves periodic orbit continuation, offering a reliable and efficient numerical tool for future cislunar mission design.

This publication has 14 references indexed in Scilit: