Deep Reinforcement Learning-Based Approach With Varying-Scale Generalization for the Earth Observation Satellite Scheduling Problem Considering Resource Consumptions and Supplements

The Earth observation satellite scheduling problem (EOSSP) becomes significantly complex when considering the real-world dynamic onboard resource environment, making it difficult to develop efficient methods with varying-scale generalization for onboard autonomous scheduling. To address this issue, this article proposes a deep reinforcement learning-based approach for the EOSSP considering resource consumptions and supplements (EOSSP-RCS). The proposed scheduling model, Transformer-based encoder–decoder architecture with temporal encoding (TRM-TE), utilizes a TRM-TE, which enhances the model's perception of time-related constraints by incorporating actual execution time information. The scheduling model is trained using the REINFORCE with critic baseline algorithm. Computational experiments show that the proposed approach achieves relatively good performance in varying-scale generalization for the EOSSP-RCS.