PowerAtlas: Towards Electricity-Computing Co-Scheduling for Power Systems

2026-07-29

AI Summary

The surge in AI workloads has transformed data centers into highly volatile, spatiotemporally flexible loads subject to stringent power grid constraints. Scheduling strategies generated by general-purpose large language models often prove infeasible due to their neglect of physical grid limitations. To address this challenge, this work proposes PowerAtlas—the first co-scheduling framework that deeply integrates large language model agents with explicit power grid physical constraints—to jointly optimize electrical and computational tasks while satisfying both grid operational rules and service-level agreements. We introduce ECBench, a benchmark comprising 2,000 instances with known optimal solutions, and demonstrate across three open-source large models that our approach consistently achieves feasible and cost-effective scheduling under real-world provincial grid data. The implementation is publicly released.

Abstract

The rapid growth of AI workloads is turning data centers into large-scale, volatile, yet spatiotemporally flexible grid loads, creating an urgent need for coordinated electricity-computing scheduling. Under stringent grid constraints, schedules from general-purpose large language models (LLMs) are often infeasible, causing line-flow violations and unserved load. We present PowerAtlas, an LLM-agent framework for electricity-computing co-scheduling that integrates historical instances, domain knowledge, and physical constraints to produce joint decisions satisfying both grid operational rules and the service-level agreements (SLAs) of computing tasks. Working with a provincial power utility in China, we built an experimental electricity-computing network and validated the decision loop on real data-center data; from de-identified operational data we further constructed ECBench, a benchmark of 2,000 scheduling instances with oracle-optimal solutions. Experiments across eleven LLMs demonstrate the effectiveness of PowerAtlas under realistic physical operating conditions, with consistent feasibility and cost gains across three open-weight backbones. Our code is publicly available at https://github.com/JAVA-Jiang/PowerAtlas.

Problem

• electricity-computing co-scheduling
• grid constraints
• data center load
• scheduling feasibility
• power system operation

Innovation

• electricity-computing co-scheduling
• LLM agent
• grid-aware scheduling
• physical constraint integration
• ECBench