摘要
随着我国航天技术的快速发展,包括导航、遥感和通信在内的航天资源越来越丰富,同时,国民经济和国防建设对航天信息的需求迫切,如何充分地应用航天信息和航天资源,成为一个新的研究内容。分析了航天信息应用的具体模式,采用深度强化学习的建模和优化方法,探索和研究了具体应用场景下的深度强化学习对应用需求的筹划和决策安排,从而在理论上验证了将人工智能方法应用于航天信息综合应用决策的可行性,为航天信息应用的大众化、平民化提供了支撑。通过仿真环境,测试在有限迭代范围内多个模型的优化速度。实验证明,在价值模型中选择DoubleDQN网络,其优化决策的收敛性能更好。
With the rapid development of China’ s aerospace technology, the space resource are becoming more and more abundant,including navigation,remote sensing,and communications. At the same time,the urgent needs of the national economy and national defense for space information,and how to make full use of aerospace information and space resources,become new research content. The specific modes of aerospace information application are analyzed, the modeling and optimization methods of deep reinforcement learning are adopted, and the planning and decision-making arrangements for the application of deep reinforcement learning are investigated under specific application scenarios. This theoretically verifies the feasibility of the artificial intelligence method applied to aerospace information integrated application decision-making, and provides support for the popularization of aerospace information applications. Through the simulation environment,the optimization speed of multiple models in a limited iteration range is tested. The experiment results show that the selection of Double DQN network in the value model has better convergence performance for optimization decision.
作者
王港
帅通
陈金勇
高峰
WANG Gang;SHUAI Tong;CHEN Jinyong;GAO Feng(CETC Key Laboratory of Aerospace Information Applications, Shijiazhuang 050081, China)
出处
《无线电工程》
2019年第7期564-570,共7页
Radio Engineering
基金
中国电子科技集团公司航天信息应用技术重点实验室开放基金资助项目(SXX18629X022)
关键词
深度强化学习
航天信息应用
DQN
需求建模
价值优化
deep reinforcement learning
aerospace information applications
demand modeling decision making
value optimization
