Collaboration among distributed agents is fundamental to many complex systems, particularly in communication networks where connectivity must be maintained under energy constraints. In this study, we utilize intelligent agents (nodes) trained through reinforcement learning techniques to establish connections with their neighbors, ultimately leading to the emergence of a large-scale communication cluster. Notably, there is no centralized administrator; instead, agents must adjust their connections based on information obtained from local observations. The connection strategy is formulated using a physical Hamiltonian, thereby categorizing this intelligent system under the paradigm of "Physics-Guided Machine Learning". Agents are trained via a Deep Q-Network using local observations to minimize changes in the Hamiltonian, enabling adaptive decision-making in dynamic environments. Simulation results demonstrate that the proposed collaborative strategy forms robust large-scale communication clusters while reducing transmission energy compared to baseline approaches. The network maintains high connectivity under agent mobility, density variations, node failures, and environmental obstacles, highlighting strong adaptability and resilience. These findings indicate that physics-guided reinforcement learning provides an effective mechanism for distributed topology optimization in emerging IoT and vehicular communication networks.

翻译：分布式智能体间的协作是许多复杂系统的基础，尤其是在需在能量约束下维持连通性的通信网络中。本研究采用通过强化学习训练的智能节点，与相邻节点建立连接，最终涌现出大规模通信集群。值得注意的是，该网络不存在集中式管理员，智能体必须依据局部观测信息调整连接策略。该连接策略采用物理哈密顿量进行建模，从而将这一智能系统纳入"物理引导机器学习"范式。智能体通过深度Q网络基于局部观测进行训练，以最小化哈密顿量的变化，实现动态环境中的自适应决策。仿真结果表明，与基线方法相比，所提出的协作策略在降低传输能耗的同时形成了稳健的大规模通信集群。该网络在节点移动、密度变化、节点失效及环境障碍条件下仍保持高连通性，展现出强大的适应性与鲁棒性。研究结果表明，物理引导的强化学习为新兴物联网及车载通信网络中的分布式拓扑优化提供了有效机制。