Research on Earth Observation Satellite Scheduling based on Deep Reinforcement Learning and Rule Method

Scheduling Earth observation satellites is a challenging multiple constraint combinatorial optimization issue. Target sequencing and establishing the observation time are the two components of Earth observation satellite scheduling. In this paper, a Markov Decision Process is used to model the satellite observation process, and the scheduling goal is to maximize the total profit rate. To implement target selection and sequencing, a scheduling decision model is built and trained by the deep reinforcement learning algorithm, and the observation time is determined using the earliest possible observation time rule approach. Experiments are conducted on randomly generated data, and the results show that the scheduling decision model is feasible and has the same effectiveness with unknown data.