Abstract: Due to the limited computing resources and very different capabilities of the space-earth integration network， the processing power of complex tasks is not strong and important tasks fail to be processed. Therefore， a three-layer computational offloading overhead model that offloads tasks to local-backbone-edge access nodes was established and the optimal offloading strategy was formulated through the optimal offloading algorithm based on DQN(deep Q-learning network). Firstly， based on the characteristics of the three types of computing nodes(offloading sites)in the network， including the space-based backbone nodes， edge access nodes and ground-based backbone nodes， the expressions of the delay， energy consumption and the corresponding expressions of different offload sites were given. Then， based on this， the DQN algorithm was proposed to complete the low-delay， low-energy offloading process. Finally， simulation results show that the DQN algorithm can improve the speed of task execution， reduce the energy consumption of terminal equipment， and effectively improve the current situation of computing node resources in the network.

Key words: space-earth integration network; computational offloading; offloading overhead model; delay; energy consumption; DQN(deep Q-learning network)