Low earth orbit satellite switching strategy based on multi-objective multi-agent reinforcement learning
To address the challenges of uneven traffic demand distribution and excessive concurrent handover among ground users in low earth orbit satellite communication systems,this paper proposes a low earth orbit satellite handover strategy based on multi-objective multi-agent collaborative deep reinforcement learning.The strategy aims to optimize the ground cell user traffic demand satisfaction,handover delay,and user conflict as the objectives,and adopts a multi-agent collaborative deep learning algorithm to optimize the objectives.Each agent is only responsible for the satellite handover strategy of one cell user,and the agents cooperate with each other by sharing rewards,thus achieving the effect of multi-objective optimization.Simulation results show that the average user traffic satisfaction of the proposed handover strategy is 73.1%,and the average handover delay is 343 ms.Compared with heuristic algorithms,the proposed strategy can better meet the traffic demand of ground cell users and balance the satellite network's load.
low earth orbit satellite networkmulti-satellite handovermulti-objective optimizationmulti-agent deep re-inforcement learning
NETL
NSTL
万方数据
维普
