【摘要】 针对工业智能体互联网中智能体协同关系与通信拓扑动态、原子化调整挑战，提出一种支持“通信-控制”协同的意图驱动控制系统架构及原子化重构机制。首先，设计了将智能体控制逻辑与通信拓扑统一封装为可动态演化的形式化协同状态模型。进一步，提出了双缓冲原子化协同状态转移机制。基于该机制，通过后台重构与原子化切换实现了智能体内部行为与外部连接的微秒级、无中断同步演化。特别地，在控制器内构建了支持动态增删的协议驱动与数据缓存池，通过“增量预热、存量延后”策略，保障多协议网络拓扑重构的数据一致性与时序确定性。实验结果表明，所提架构可根据视觉协同意图，在17.4μs内完成协同状态切换，实现同步误差<4.1 ms的动态协同分拣任务，证明了该架构可有效支撑智能体网络高效协同与按需重构。更多还原

【Abstract】 To address the challenges of dynamic and atomic adjustment of cooperative relationships and communication topologies in the industrial Internet of intelligent agent, an intent-driven architecture supporting “communication control” collaboration was proposed. Firstly, a formalized, dynamically evolvable cooperative state model was designed to unify agent control logic and communication. Further, a double-buffer atomic transition mechanism enabled agents’ internal behaviors and external connections to evolve synchronously in microseconds without interruption. Additionally, a dynamic protocol-driven data cache pool within the controller used an “incremental preheating and stock delay” strategy to guarantee data consistency and timing determinacy during network reconfiguration. Experimental results show the architecture completes cooperative state switching within 17.4 μs and achieves a dynamic sorting task with a synchronization error under 4.1 ms, effectively supporting efficient collaboration and on-demand reconfiguration of agent networks.更多还原