伴随科技的快速发展，多无人船系统在军事、营救以及护送等任务场景中展现出巨大的潜力。研究旨在基于多智能体深度强化学习算法，对多无人水面航行器系统的编队构造问题进行探究。针对传统多智能体深度确定性策略梯度算法（MADDPG）收敛速度较慢的问题，研究通过在值函数阶段引入了注意力机制来提升多无人水面航行器系统的编队决策模型的收敛速度，并通过无人水面航行器的编队模型与编队避碰和编队构造奖励函数的配合，提升了多无人水面航行器完成编队构造任务的效率。最后，通过仿真结果验证了本文所提出的方法可有效地完成多种环境下的多无人水面航行器编队构造任务，为未来多无人船编队构造应用提供理论研究基础。

With the rapid development of science and technology, multi-unmanned ship systems have shown great potential in military, rescue and escort mission scenarios. The purpose of this paper is to explore the formation construction problem of multiple unmanned surface vehicle systems based on multi-agent deep reinforcement learning algorithm. Considering the sluggish convergence speed of the conventional multi-agent deep deterministic policy gradient algorithm (MADDPG), this study incorporates the attention mechanism into the value function stage to enhance the convergence speed of the formation decision model for a multi-UAV system. Through the cooperation of the formation model of the unmanned surface vehicle with the formation collision avoidance and the formation construction reward function, the efficiency of the multi-UAV to complete the formation construction task is finally improved. The simulation results conclusively demonstrate the efficacy of the proposed method in accomplishing multi-unmanned surface vehicle formation construction tasks, thereby establishing a solid theoretical foundation for future applications of multi-unmanned ship formation construction.