关键词: 任务规划, 多目标序列规划, 并行计算, 计算统一设备架构

Abstract:

Aiming at the satellite sequence planning problem of continuous rendezvous with multiple noncoplanar targets, a fast planning method based on compute unified device architecture （CUDA） is proposed. The rendezvous timing is taken as the moment when the target crosses the orbital plane of the satellite. First, a set of candidate solutions for the rendezvous sequence is formed, and the calculation of the transfer velocity increment between targets based on the orbital dynamics model is simplified to a bilinear interpolation problem based on the transfer velocity and the flight angle. Then, texture memory is applied for fast estimation of the velocity increment, the parallelization at the level of candidate plan is realized by multiple thread blocks, and the parallelization of the transfer velocity increment calculation between targets and the accumulation of the total velocity increment are realized by parallel reduction and shared memory within a single candidate plan. Finally, an exhaustive search optimization approach is applied to the candidate solutions and the multi-target rendezvous sequence planning for minimizing the velocity increment is achieved. Simulation results show that this method can quickly plan the optimal rendezvous sequence, which is more than 20 times faster than serial calculation. As intelligent computing power gradually becomes more widespread in space, it is expected to significantly improve the efficiency of on-orbit planning for scenarios such as multi-target observation in the future.

Key words: mission planning, multiple-target sequence planning, parallel computing, compute unified device architecture (CUDA)