图机器学习:电力系统的新机遇
文章背景与核心概要
现代电力系统正经历着前所未有的运行复杂性,这主要由能源基础设施的去中心化、可再生能源的大规模并网以及多时间尺度实时决策的需求所驱动。尽管传统的基于模型的物理方法保持了高准确性,但它们往往难以满足现代电网运行严苛的时延要求。为了弥补这一鸿沟,机器学习(ML)作为一种数据驱动的替代方案应运而生。由于电网拓扑结构从根本上决定了电力系统的行为,图机器学习(GML)通过无缝嵌入拓扑依赖作为归纳偏置(inductive bias),成为了一个自然的建模范式。
本文综述了近 800 篇处于 GML 与电力系统交叉领域的学术研究,评估了其核心应用,包括负荷/功率预测、状态估计、优化与控制、故障诊断以及网络安全。尽管该领域发文量呈指数级增长,但也面临着现实部署有限、可解释性缺失以及缺乏标准化基准等严峻挑战。为此,作者建立了一个面向 ML 就绪型电网基准的结构化需求目录,旨在指导未来的数据集创建,并提升全行业研究的科学严谨性。
Metadata & Reference Information
- arXiv ID: arXiv:2608.16494 [cs.LG]
- Subjects: Machine Learning (
cs.LG); Artificial Intelligence (cs.AI); Computational Engineering, Finance, and Science (cs.CE); Systems and Control (eess.SY) - DOI: 10.48550/arXiv.2608.16494
- Submission Date: August 17, 2026
- License: Creative Commons Attribution 4.0 International

元数据与参考信息 * arXiv ID: arXiv:2608.16494 [cs.LG] * 研究学科: 机器学习 (
cs.LG);人工智能 (cs.AI);计算工程、金融与科学 (cs.CE);系统与控制 (eess.SY) * DOI: 10.48550/arXiv.2608.16494 * 提交日期: 2026年8月17日 * 许可证: 知识共享署名 4.0 国际许可协议
Authors
- Martin Sadric
- Sebastian Pütz
- Christian Nauck
- Veit Hagenmeyer
- Frank Hellmann
- Dirk Witthaut
- Benjamin Schäfer
作者 * Martin Sadric * Sebastian Pütz * Christian Nauck * Veit Hagenmeyer * Frank Hellmann * Dirk Witthaut * Benjamin Schäfer
Executive Summary
Modern power systems are experiencing escalating operational complexities driven by the decentralization of energy infrastructure, the integration of renewable energy sources, and the necessity for real-time decision-making across multiple timescales. While traditional model-based approaches maintain high accuracy, they often struggle to meet the strict latency demands of modern grid operations.
执行摘要 在能源基础设施去中心化、可再生能源并网以及跨多时间尺度实时决策需求的驱动下,现代电力系统正面临日益严峻的运行复杂性。虽然传统的基于模型的方法能够保持高准确度,但它们往往难以满足现代电网运行对时延的严格要求。
To bridge this gap, Machine Learning (ML) has emerged as a data-driven alternative. Because grid topology fundamentally dictates power system behavior, Graph Machine Learning (GML) serves as a natural modeling paradigm by seamlessly embedding topological dependencies as an inductive bias.
为了弥补这一鸿沟,机器学习(ML)作为一种数据驱动的替代方案应运而生。由于电网拓扑结构从根本上决定了电力系统的行为,图机器学习(GML)通过无缝地将拓扑依赖关系嵌入为归纳偏置,提供了一种天然的建模范式。
This paper surveys nearly 800 studies at the intersection of GML and power systems, evaluating core applications including: * Forecasting * State estimation * Optimization and control * Fault diagnosis and cybersecurity
本文综述了近 800 篇处于 GML 与电力系统交叉领域的学术研究,评估了其核心应用,包括: * 预测 * 状态估计 * 优化与控制 * 故障诊断与网络安全
Key Findings & Insights
Why Power Systems Suit GML
Power systems present a uniquely rigorous and information-rich benchmark environment for graph-based machine learning because they combine: * Hard physical constraints (e.g., power flow equations) * Complex multi-scale dynamics * Stringent safety-critical operational requirements * High-dimensional domains characterized by scarce labeled data
关键发现与洞察
为什么电力系统非常适合图机器学习(GML)
电力系统为基于图的机器学习提供了一个极其严苛且信息丰富的基准测试环境,因为它们兼具以下特征: * 严格的物理约束(例如潮流方程) * 复杂的多尺度动态特性 * 严苛的安全关键型运行要求 * 以稀缺标记数据为特征的高维领域
Conversely, classical power system solvers greatly benefit from GML by gaining scalable, topology-aware approximations that offer superior computational efficiency and generalization capabilities.
相反,传统的电力系统求解器也能从 GML 中获益匪浅,从而获得具备可扩展性、拓扑感知能力的近似解,进而提供卓越的计算效率和泛化能力。
Open Challenges & Limitations
Despite rapid publication growth, the field faces substantial hurdles: * Limited Real-World Deployment: Few models have successfully transitioned to live operational environments, largely due to safety and trust concerns. * The Need for Interpretability: Safety-critical domains require interpretable ML decisions, which remain difficult to extract from standard black-box architectures. * Reproducibility Crisis: Standardized benchmarks and open datasets are critically scarce, rendering many published results difficult to reproduce and undermining long-term scientific credibility.
公开挑战与局限性
尽管相关论文数量快速增长,该领域仍面临重大障碍: * 现实世界部署有限: 极少有模型成功过渡到实际运行环境中,这主要归咎于安全和信任方面的顾虑。 * 对可解释性的需求: 安全关键领域要求机器学习的决策具备可解释性,而这从标准的黑盒架构中很难提取。 * 可复现性危机: 标准化基准和开放数据集极度匮乏,导致许多已发表的研究成果难以复现,并削弱了长期的科学可信度。
Call to Action
The authors establish a structured requirements catalog for ML-ready power grid benchmarks designed to steer future dataset creation and elevate scientific rigor across studies. The research community is urged to prioritize dedicated benchmark development alongside the open-access release of underlying datasets and models.
行动倡议
作者针对“面向机器学习的电网基准”建立了一个结构化的需求目录,旨在指导未来的数据集创建,并提升各项研究的科学严谨性。作者强烈呼吁研究界在推进专属基准开发的同时,实现底层数据集与模型的开源开放。