一种动态锁定式多星协同任务规划的A2C算法

An A2C Algorithm for Dynamic Locking-based Multi-satellite Collaborative Task Planning

提出了一种基于强化学习算法的多星协同任务规划方法。该方法将多星协同任务规划视为多个双星协同规划,采用相邻双星锁定的方式来共享双星任务规划结果信息,基于A2C(优势动作评价)强化学习算法对双星任务规划的结果进行再调整。针对多星协同任务规划状态空间复杂且变化的问题,设计了两级状态空间和价值评价函数决策的方法,对强化学习所依赖的状态空间进行维度限制,确保智能体对任务进行调整的过程不影响状态空间维度。算法设计过程考虑了多种约束条件,设置了天气、成像质量和成像优先级等可调参数作为强化学习A2C算法的评价参数,这些可调参数有助于用户自定义决策评价体系。最后,通过仿真验证了算法的可行性。仿真结果表明,该算法多星协同任务规划的组合任务抛弃率小于10%。

A multi-satellite collaborative task planning method based on reinforcement learning algorithm is proposed.This method regards multi-satellite collaborative task planning as multiple dual-satellite collaborative plannings,and uses adjacent dual-satellite locking to share task planning result information.Based on the A2C(advantage actor-critic) reinforcement learning algorithm,the results of dual-satellite task planning are readjusted.To address the complex and changing state space of multi-satellite collaborative task planning,a two-level state space and value evaluation function decision-making method is designed to limit the dimensionality of the state space relied on by reinforcement learning,ensuring that the intelligent agent's task adjustment process does not affect the dimensionality of the state space.In the process of algorithm design,various constraints are considered,and adjustable parameters such as weather,imaging quality,and imaging priority are set as evaluation parameters for the A2C algorithm of reinforcement learning.These adjustable parameters help users customize the decision-making evaluation system.Finally,simulations are conducted to validate the feasibility of the algorithm.The simulation results show that the combination task abandonment rate of the algorithm for multi-satellite collaborative task planning is less than 10%.