EOS Autonomous Mission Scheduling with Orbital Threat Avoidance

This paper investigates the autonomous scheduling problem for Earth observation satellite (EOS) imaging missions under orbital threats. First, within the increasingly complex space environment, typical scenarios of EOS encountering orbital threats are analyzed. Based on the urgency level of threats and their avoidance deadlines, a scheduling optimization model is established to minimize the impact of avoidance maneuvers on imaging mission execution. Then, a reinforcement learning (RL)-based self-learning genetic algorithm (GA) is proposed. In GA, decision variables of mission are encoded as the chromosomes. The solution space is effectively explored through the crossover, mutation, and selection operations of the GA to approximate the optimal solution. The performance of GA is seriously affected by the probabilities of crossover and mutation that are difficult to tune manually. To address this problem, an RL-based framework is introduced to adaptively adjust these parameters. The state representation, action selection strategy, and reward function are designed within the framework. Additionally, a two-stage learning strategy intergrating state-action-reward-state-action (SARSA) and Q-learning algorithm is proposed via predefined transition conditions. Finally, simulation results demonstrate the algorithm’s effectiveness in balancing the conflicts between imaging and avoidance missions. The algorithm successfully achieves a trade-off between the two types of mission.