With the rapid development of space and information fi eld, hot content distribution intensive scenes will become one of the key directions of satellite network application, and satellite content delivery network (CDN) network is an important means to improve the effi ciency of air and space content distribution.In the architecture of satellite CDN network, due to the uneven time and space of business requirements, the scarcity of satellite resources and the insuffi cient adaptability of existing scheduling algorithms, scheduling algorithms for satellite resources are faced with problems such as high resource dimension, many computing states and large amount of computation, which will reduce the accuracy, response speed and computing performance of scheduling decisions.To solve this problem, a resource scheduling algorithm based on Deep Q-Learning (DQN) algorithm was proposed to improved the effi ciency and accuracy of satellite resource scheduling, and intelligently and quickly perceived the resource situation to make scheduling decisions.Firstly, the user requests were classifi ed, and the shortest path set that the satellite could communicated with was calculated according to the time-varying trajectory of the satellite and the resources of the satellite and the ground.After that, the related information of satellites and users was quantifi ed by Markov model modeling, and the optimal CDN storage node of satellites was calculated by DQN algorithm, which achieved the eff ects of reduced user request delay, reduced satellite-ground resource occupancy rate and improved cache hit rate.
