Abstract: To address the problems of high energy consumption and latency in task offloading,as well as the inefficient allocation of cloud-edge-end resources,a dual-agent dynamic offloading strategy named DQN-NA that integrates Deep Q-Network (DON) with Nash equilibrium (NA) under a cloud-edge-end collaborative framework was proposed.By formulating a reward function based on both delay and energy consumption and introducing a dual-agent game-theoretic model,a decision-making architecture was developed to dynamically select optimal offloading strategies.The method enabled adaptive offloading by reasonably distributing computing resources among local devices,edge servers and the cloud,thereby effectively reducing task processing delay and energy consumption and achieving an optimal trade-off within acceptable limits.The proposed strategy was validated on a steel manufacturing production line,where the straightening machine serves as a critical process.Experimental results showed that compared with traditional offloading approaches,the DQN-NA strategy significantly reduced both delay and energy consumption while achieving a better balance between them,which fully demonstrated the applicability and robustness of the method in complex industrial environments.

Key words: edge computing, computation offloading, dual-agent system, Nash equilibrium, steel production line

摘要： 针对任务计算卸载存在的高能耗与高时延问题,以及云、边、端资源分配不合理问题,提出一种基于云、边、端协同的深度Q网络与纳什均衡相结合的双智能体动态卸载策略(DQN-NA)。通过建立时延能耗二者的奖励函数优化问题,并引入双智能体博弈建模,设计出了动态选择卸载策略的决策架构,实现了任务在本地、边缘和云端之间合理分配资源的自适应卸载策略,有效降低了处理任务的时延和能耗,使得时延和能耗在任务处理允许的范围内达到最优状态。所提方法在以拉矫机为关键工序的钢铁制造产线进行了验证,结果显示,与传统的计算卸载策略相比,DQN-NA卸载策略既有效平衡了时延和能耗,也使得时延和能耗得到了显著降低,充分证明了该方法在复杂工业环境中的适用性和广泛性。

关键词: 边缘计算, 计算卸载, 双智能体, 纳什均衡, 钢铁制造产线