基于深度强化学习的卫星动态调度策略  

Satellite dynamic scheduling strategy based on deep reinforcement learning

作　　者：赵昊天 刘越 杨鹏 ZHAO Haotian;LIU Yue;YANG Peng(School of Computer Science and Engineering,Tianjin University of Technology,Tianjin 300384,China)

机构地区：[1]天津理工大学计算机科学与工程学院,天津300384

出　　处：《天津理工大学学报》2024年第3期91-96,共6页Journal of Tianjin University of Technology

基　　金：复杂系统仿真国防科技重点实验室基金(DXZT-JC-ZZ-2019-010)。

摘　　要：为了解决卫星运动过程中对于新增任务的资源分配难问题,提出了将卫星调度模型与深度强化学习结合起来的卫星调度优化框架。以北方某卫星任务站的任务调度过程作为基础,使用Agent仿真技术建立卫星任务调度模型,模型中引入深度Q网络(deep Q-network,DQN)算法。调度模型根据地面任务站的实时数据对卫星的使用情况进行调整,将调整过后模型的数据做成数据集,用于深度学习的训练。训练成熟的网络与地面站任务调度结合起来,用于支持卫星实时动态调度。实验表明引入DQN可以很好地优化卫星调度,比现有优化方法更加高效。In order to solve the difficult problem of resource allocation for new tasks in the process of satellite movement,a satellite scheduling optimization framework combining satellite scheduling model and deep reinforcement learning is proposed.Based on the task scheduling process of a satellite task station in the north,the Agent simulation technology is used to establish a satellite task scheduling model,and the deep Q-network(DQN)algorithm is introduced into the model.The scheduling model adjusts the usage of the satellite according to the real-time data of the ground mission station,and the adjusted data of the model is made into a data set for deep learning training.A well-trained network is combined with ground station task scheduling to support real-time dynamic scheduling of satellites.Experiments show that the introduction of DQN algorithm can optimize satellite scheduling well,which is more efficient than the existing optimization methods.

关 键 词：在线仿真 卫星调度 深度强化学习 随机任务 实时优化 

分 类 号：TP391.9[自动化与计算机技术—计算机应用技术]