With the continuous development and innovation of technology, the application of Unmanned Aerial Vehicles (UAVs) in the military field is becoming increasingly widespread. In particular, the use of multiple UAVs to form a swarm for combat has become a hot topic of research in various countries. As battlefield missions become more complex, the intelligence of UAV control is an inevitable trend. To address the problem of autonomous rendezvous for large-scale UAV swarms, the Deep Deterministic Policy Gradient (DDPG) algorithm that combined with mean field game theory and partial observability constraints are introduced. By incorporating the “centralized training, distributed execution” mechanism, the Partially Observable Mean Field Deep Deterministic Policy Gradient (PO-MFDDPG) algorithm is proposed. The simulation results show that the performance of the PO-MFDDPG algorithm is superior to the previous improved algorithms.