Abstract: In the mobile edge computing scenario, the task cache delay becomes excessively long due to the rapid growth of data traffic. To address this issue, a cache scheme based on deep reinforcement learning algorithm in mobile edge computing scenarios is proposed. Firstly, under a 5G-based space-air-ground integrated network (SAGIN) architecture, a cache and communication model is established to minimize the content cache delay. Secondly, the SAC (soft actor critical) algorithm is used to interfere with the local minimum cache delay, and a new scheme is accepted with a certain probability, thereby achieving the global maximum cache hit rate. Finally, the above process is iterated repeatedly to obtain an optimal solution for the target problem, ensuring that the task files are pre-delayed in the optimal location. The simulation results show that under the SAGIN cooperation architecture, compared with the PPO (proximal policy optimization) scheme, the cache scheme can reduce the optimization transmission efficiency, reducing the cache delay by 5.30%, and improving the cache hit rate by 3.90%.