The proliferation of large-scale artificial intelligence and data-intensive applications has spurred the development of Computing Power Networks (CPNs), which promise to deliver ubiquitous and on-demand computational resources. However, the immense energy consumption of these networks poses a significant sustainability challenge. Simultaneously, power grids are grappling with the instability introduced by the high penetration of intermittent renewable energy sources (RES). This paper addresses these dual challenges through a novel Two-Stage Co-Optimization (TSCO) framework that synergistically manages power system dispatch and CPN task scheduling to achieve low-carbon operations. The framework decomposes the complex, large-scale problem into a day-ahead stochastic unit commitment (SUC) stage and a real-time operational stage. The former is solved using Benders decomposition for computational tractability, while in the latter, economic dispatch of generation assets is coupled with an adaptive CPN task scheduling managed by a Deep Reinforcement Learning (DRL) agent. This agent makes intelligent, carbon-aware decisions by responding to dynamic grid conditions, including real-time electricity prices and marginal carbon intensity. Through extensive simulations on an IEEE 30-bus system integrated with a CPN, the TSCO framework is shown to significantly outperform baseline approaches. Results demonstrate that the proposed framework reduces total carbon emissions and operational costs, while simultaneously decreasing RES curtailment by more than 60% and maintaining stringent Quality of Service (QoS) for computational tasks.

翻译：大规模人工智能与数据密集型应用的激增推动了算力网络（CPN）的发展，该网络旨在提供无处不在、按需分配的计算资源。然而，这些网络巨大的能源消耗带来了严峻的可持续性挑战。与此同时，电力系统正面临间歇性可再生能源（RES）高渗透率所引入的不稳定性问题。本文通过一种新颖的两阶段协同优化（TSCO）框架应对上述双重挑战，该框架协同管理电力系统调度与CPN任务调度，以实现低碳运行。该框架将复杂的大规模问题分解为日前随机机组组合（SUC）阶段和实时运行阶段。前者采用Benders分解法求解以保证计算可处理性，而在后者中，发电资产的经济调度与由深度强化学习（DRL）智能体管理的自适应CPN任务调度相耦合。该智能体通过响应动态电网条件（包括实时电价和边际碳强度）做出智能的、具备碳感知能力的决策。通过在集成CPN的IEEE 30节点系统上进行大量仿真，结果表明TSCO框架显著优于基准方法。仿真结果证明，所提框架在降低总碳排放和运行成本的同时，将可再生能源弃电量减少了60%以上，并保持了计算任务严格的服务质量（QoS）。