This paper investigates the multi-agent formation control problem for underactuated unmanned surface vehicles (USVs) and proposes a distributed control strategy based on partial differential equations (PDE), integrated with an improved artificial potential field (APF) method for obstacle avoidance. A leader-follower formation framework is first established, where the desired positions and tracking velocities of follower vessels are derived. Subsequently, a PDE-based formation control law is designed to ensure stability in multi-USV cooperative motion. The APF approach is enhanced by introducing tangential force components and angle-adaptive coefficients to optimize obstacle avoidance trajectories. The proposed algorithm is validated through simulations using ArduPilot-SITL, ROS, and QGroundControl, demonstrating its effectiveness in formation maintenance, formation switching, and obstacle avoidance. Experimental results show that the method achieves rapid convergence of formation errors while effectively avoiding static obstacles, exhibiting strong robustness and practical applicability.