Time-of-Flight Estimation in Low-Thrust Transfers between Asteroids Using Deep Neural Networks

This paper focuses on optimizing the transfer time for multiple asteroid rendezvous missions utilizing low-thrust propulsion. An indirect method based on Pontryagin’s maximum principle is employed to generate training data by optimizing low-thrust trajectories for minimal time of flight. A deep neural network (DNN) is subsequently trained to estimate the minimum time of flight (TOF) between each pair of asteroids. To enhance the accuracy of predictions for rapid transfers (less than 150 days), the dataset is augmented with additional trajectory simulations. Leveraging the trained neural network, a Beam Search (BS) algorithm is implemented to efficiently determine the fastest sequence of rendezvous with multiple asteroids, starting from an arbitrary initial asteroid. The results demonstrate that an optimal rendezvous sequence involving 15 asteroids is achieved, with an average time of flight of 107.5 days per segment.