Satellite Formation Configuration Design Using Physics-Informed Neural Networks

A physics-informed-neural-network-based method for satellite formation configuration design is proposed, which overcomes the high computational complexity and dependence on initial guesses inherent to traditional nonlinear programming approaches. Formation parameters (relative eccentricity vector and relative inclination vector) are encoded as the neural network’s outputs, while mission constraints (collision avoidance and communication-range limits) and the optimization objective (safety margin) are transformed into physics-based penalty terms in the loss function, enabling training without any dataset. Simulation experiments verify the physical consistency of the proposed method under complex constraints.