Journal of Systems Engineering and Electronics ›› 2026, Vol. 37 ›› Issue (4): 1341-1353.doi: 10.23919/JSEE.2026.000142

• SYSTEMS ENGINEERING • Previous Articles    

Reinforcement survivable routing for resource-limited LEO satellite network

Yuanyuan Nie1(), Zhigeng Fang2,*(), Su Gao3(), Jiajia Cai4()   

  1. 1China Astronautics Standards Institute, Beijing 100071, China
    2School of Economics and Management, Nanjing University of Aeronautics and Astronautics, Nanjing 210016, China
    3China Academy of Space Technology, Beijing 100094, China
    4School of Business, Yangzhou University, Yangzhou 225127, China
  • Received:2024-03-06 Online:2026-08-18 Published:2026-09-03
  • Contact: Zhigeng Fang E-mail:yynnuaa@163.com;zhigengfang@163.com;su_gao06@163.com;cai_jiajia@yeat.net
  • Supported by:
    This work was supported by the youth program of humanities and social sciences research in the Ministry of National Education of China (23YJCZH180). It is also supported by the Natural Science Foundation of the Jiangsu Higher Education Institutions of China (23KJD120003).

Abstract:

In order to provide low Earth orbit (LEO) satellite network survivability with limited resources, and effectively respond to the risk of on orbit operation, a survivable routing algorithm based on multi-agent reinforcement learning is proposed. The reinforcement survivable routing establishes the reward function from the perspective of network utility, and describes the improvement of survivability in the failure scenario as a joint optimization problem, which solves the online routing decision-making problems after LEO satellite network failure. Each cognitive user agent and satellite agent can independently determine the routing strategy for communication missions according to the time slot topology, node survival status, storage space and other environment information of the satellite network. The simulation results show that compared to the weighted semi-distributed routing algorithm (WSDRA) and Fibonacci multipath load balancing (FMLB) algorithm, the proposed multi-agent double deep Q-learning network (MADDQN) reinforcement survivable routing algorithm can adapt to satellite failure and topology changes, and improve network utility by 14% and 9% respectively in human attack scenario, and by 17% and 13% respectively in random failure scenario.

Key words: low Earth orbit (LEO) satellite network, survivable routing, reinforcement learning, network utility