Abstract: In the Computing Power Network (CPN), User Devices (UE) with limited computational capability and battery capacity rely on external computing nodes for task processing. Existing studies mainly focus on direct Workflow Task (WT) offloading, neglecting the UE of waiting delay and high energy consumption caused by task dependencies. When data from preceding tasks is cached on UE, prolonged waiting not only increases the load on computing nodes but also forces UE into high-energy consumption states, posing significant challenges for optimizing energy-sensitive tasks. To address these challenges, a dynamic optimization and offloading (DOOWT) mechanism was developed to improve energy efficiency. The Workflow Structure Optimization algorithm (WSO) was utilized to rearrange the task graph, reducing inter-task waiting delays and lowering overall energy consumption. Additionally, a Dependency-Based Task Offloading (WTO) algorithm leveraging Deep Deterministic Policy Gradient (DDPG) was employed to dynamically adjust offloading strategies, enhancing computational performance and resource utilization in CPNs. Experimental results demonstrated that, compared with conventional methods, DOOWT reduces task waiting delay by 60%, shortens the average completion time of WT by 79%, and decreases overall energy consumption by 82%. These findings provide theoretical and technical support for the efficient optimization and scheduling of energy-sensitive tasks.

Key words: Computing Power Network (CPN), dependency task offloading, Deep Reinforcement Learning (DRL), energy consumption optimization, task offloading

摘要： 在算力网络(CPN)中，用户设备(UE)因计算能力和电源容量受限，需要依赖外部算力节点协同处理任务。现有研究多集中于直接卸载工作流任务，未充分考虑因任务依赖性导致的等待时延和高能耗问题。当前驱任务数据需在用户设备上缓存时，长时间的等待不仅增加算力节点的负载，还使UE长时间处于高功耗状态，进一步提高了能耗敏感型任务优化的复杂性。为解决上述问题，提出了一种基于动态优化与卸载机制(DOOWT)的节能优化方法。通过工作流结构优化算法(WSO)，对任务图进行重排以减少任务间的等待时延，从而降低整体能耗。结合基于DDPG的动态任务卸载算法(WTO)，实时调整卸载策略，有效提升算力网络的计算性能和资源利用效率。实验结果表明，与传统方法相比，任务等待时延减少60%，工作流任务的平均完成时间缩短79%，整体能耗降低82%。该方法为能耗敏感型任务的优化与调度提供了理论支持与技术参考。

关键词: 算力网络, 依赖任务卸载, 深度强化学习, 能耗优化, 任务卸载