The invention discloses a multi-
satellite beam hopping method based on double-layer load balancing, and relates to the technical field of low-
orbit satellite communication. In a multi-
satellite hopping beam communication scene, users are clustered based on geographic positions and service requirements of the users, and satellites capable of being served of each cluster are calculated based on elevation angles. And then, in each
resource allocation time slot, a multi-satellite hopping beam decision is made by using a deep
reinforcement learning network. Calculating a transmission data volume according to the current multi-satellite-hop beam decision, updating a user
queue, and counting the maximum time
delay of a user data packet; and the deep
reinforcement learning network is updated by taking minimization of the maximum time
delay of the user as an optimization target. And finally, based on the updated deep
reinforcement learning network, making a decision on the hopping
beam pattern and user
resource block allocation at the next moment, and circulating until the time reaches a time window, thereby realizing
dynamic balance of loads between satellites in the process. According to the method, communication resources are allocated on the user level on the basis of personalized demands of users, and meanwhile, satellite-level
dynamic load balancing is realized.