A Neural Network Approach for Generating Libration Point Orbits

In the circular restricted three-body problem, families of periodic and quasi-periodic orbits exist in the vicinity of the libration points and are of significant interest for practical applications. This work presents a neural network-based algorithm for orbit construction and station-keeping, applicable to both periodic and quasi-periodic orbits near collinear libration points. The method employs a fully connected neural network trained on labeled trajectory data and iteratively refined using trajectories generated during previous stages of the algorithm. The proposed approach was applied to orbit generation near the Sun-Earth L1 libration point and demonstrated the ability to generalize to a broader class of trajectories than those included in the training set.