A data-knowledge joint-driven reinforcement learning algorithm based on guided policy and state-prediction for satellite continuous-thrust tracking

• Published in: Advances in space research 2024-10, Vol.74 (8), p.4089-4108
• Main Authors: Wang, Xiao, Li, Jiake, Cao, Lu, Ran, Dechao, Ji, Mingjiang, Sun, Kewu, Han, Yuying, Ma, Zhe
• Format: Article
• Language: English
• Subjects: Data-driven; Knowledge-driven; Reinforcement learning; Satellite tracking
• ISSN: 0273-1177
• DOI: 10.1016/j.asr.2024.06.070

With the advancement of artificial intelligence technology, the field of space satellite tracking is also progressing towards intelligent systems. However, directly incorporating a purely data-driven learning approach into a satellite control system not only exhibits low efficiency but also necessitates a substantial number of interactive samples. To address these limitations, we propose an innovative data-knowledge joint-driven reinforcement learning algorithm based on guided policy and state-prediction (GP-SPRL) for continuous-thrust satellite tracking problems. By assuming linear quadratic tracking cost, GP-SPRL leverages orbital dynamics to facilitate an intelligent learning process and enables bidirectional guidance and update through knowledge-driven and data-driven approaches. Furthermore, to mitigate high interaction costs, GP-SPRL employs real samples to generate predicted states using Gaussian processes, thereby reducing the need for extensive interactions. Simulation results validate the efficacy of joint-driven in GP-SPRL by enhancing learning efficiency and minimizing reliance on real samples through predicted state utilization.