To address the safety issues faced by high-speed UAV swarms during formation flight in densely obstructed areas, we propose an intelligent swarm formation control framework that combines constrained A* path planning with deep reinforcement learning. A high-speed UAV swarm model is established, and an obstacle model considering the volume of the swarm is designed. Continuous planned flight trajectories are obtained by truncating path points. A reward function is designed based on the relative position and velocity relationships between UAVs, between UAVs and the target, and between UAVs and obstacles. The lower-level distributed formation intelligent controller is trained using this reward function. Simulation experiments demonstrate the effectiveness of this hierarchical control framework.