Deep reinforcement learning-based autonomous mission planning method for high and low orbit multiple agile Earth observing satellites

Concerning the onboard autonomous mission planning problem of high and low orbiting agile Earth observation satellites (AEOSs), which requires high algorithm timeliness, a deep reinforcement learning (DRL) based algorithm, namely, the multi-satellite mission planning (MSMP) algorithm, is proposed. The algorithm uses neural networks with an ‘encoder-decoder’ structure and designs a mechanism that enables each satellite to select requests in turn. These two designs allow the algorithm to achieve simultaneous scheduling of multiple satellites. After that, the MSMP uses a REINFORCE with Critic Baseline algorithm to optimize its selection strategy. Computational experiments show that the proposed algorithm can reduce the reasoning time by a factor of tens compared to the adaptive task assignment large neighborhood search (A-ALNS) algorithm, while being able to keep the revenue rate difference to A-ALNS below 5%.