Research On Edge Resource Allocation Of Multi-layer Satellite Network Based On Reinforcement Learning

With the gradual increase in the number of Earth observation satellites in orbit in my country,the substantial enhancement of observation capabilities,and the increasingly stringent requirements of users for observation quality,the "high-medium-low" integrated multi-layer satellite communication system has become a The future development direction of space-based information network.However,traditional observation satellites often need to transmit a large amount of observation data to ground users,and the energy consumption,bandwidth occupation and transmission delay caused by redundant data far exceed the system’s ability and user’s tolerance.Multi-layer satellite network edge computing architecture(MSNEC),by deploying edge servers with certain computing and caching capabilities on MEO satellites,and sinking data processing capabilities to the edge of the satellite network,thereby greatly reducing the amount of data backhaul and reducing mission response time.extension.Although MSNEC has strong application potential,the complexity of satellite networks and the scarcity of on-board resources make data offloading decision extremely difficult.Based on this,a data offloading decision-making method for multi-layer satellite networks(DQN-ESD).At the same time,exposed wireless links bring huge security risks.Considering that it is difficult to unilaterally ensure system security or mission response in practical applications to meet user needs,an intelligent encryption decisionmaking oriented to multi-layer satellite network space autonomous domain is proposed.method(DQN-AED).The specific research work of the paper is as follows:The main research contents of this paper are as follows:(1)Aiming at the problems that traditional observation satellites need to transmit a large amount of redundant data back to the ground cloud computing center for processing,which takes up a lot of bandwidth and energy,and causes large transmission delay and poor service quality,a multi-layer satellite network edge computing architecture is proposed..The concept of edge computing is introduced into the multi-layer satellite network,and the computing and storage capabilities of the cloud computing center are sunk to the edge.improves system efficiency.(2)In view of the scarcity of resources,complex network structure and high dynamics in the edge computing architecture of multi-layer satellite networks,which lead to unbalanced load and difficult selection of data offloading strategies,a data offloading decision-making method for multi-layer satellite networks is proposed.Considering factors such as the amount of observation satellite mission data,channel conditions,and edge node processing capabilities,the deep reinforcement learning algorithm is used to learn independently from historical experience,and the optimal data offloading strategy in this scenario is obtained by fitting,and the optimal link is obtained.Planning,make full use of on-board resources,and minimize the average return delay of many observation missions.The simulation results show that the proposed method is reasonable and feasible,has certain generalization ability,and has better performance than several heuristic methods.(3)Aiming at the problem that wireless links in satellite networks are scattered in free space and data are easy to be stolen and tampered with,an intelligent encryption decisionmaking method for the spatial autonomous region of multi-layer satellite networks is proposed.Relying on the local autonomous region of multi-layer satellite space,it provides secure encryption decision-making support for the return of data obtained by observation satellites.The control node in the domain fully considers the user’s requirements for the return delay of the observation task.On the basis of quantifying the security strength of the encryption strategy,the deep enhancement algorithm is used to intelligently select different encryption methods,and make a trade-off between security and delay,so as to ensure that the observation task maximizes its security on the basis of meeting the requirements of return delay.The simulation results show that the proposed method is feasible and reasonable,and has higher advantages in terms of security and practicability compared with many common comparison methods.