Self-healing mechanism of green computing power network based on GNN and multi-agent reinforcement learning

ZHANG Jianrong

doi:10.11959/j.issn.1000-0801.DXKX260388

Chinese

您当前的位置：

首页 >

文章列表页 >

Self-healing mechanism of green computing power network based on GNN and multi-agent reinforcement learning

更新时间：2026-09-15

- Self-healing mechanism of green computing power network based on GNN and multi-agent reinforcement learning
- Telecommunications Science (2026)
- 作者机构：
  
  中讯邮电咨询设计院有限公司，北京 100048
- 作者简介：
- 基金信息：
- DOI：10.11959/j.issn.1000-0801.DXKX260388
  CLC： TP393
- Received：09 June 2026，
  
  Revised：2026-07-27，
  
  Accepted：20 August 2026，
- 稿件说明：
移动端阅览
ZHANG Jianrong. Self-healing mechanism of green computing power network based on GNN and multi-agent reinforcement learning[J/OL]. Telecommunications Science, 2026.
DOI：

ZHANG Jianrong. Self-healing mechanism of green computing power network based on GNN and multi-agent reinforcement learning[J/OL]. Telecommunications Science, 2026. DOI： 10.11959/j.issn.1000-0801.DXKX260388.

摘要

绿色算力网络要求故障自愈同时兼顾恢复速度与能耗效率，现有方法多采用集中式控制，缺乏拓扑感知和分布式协作能力，且较少考虑能耗优化。为此，提出了一种基于图神经网络（Graph Neural Network，GNN）与多智能体深度确定性策略梯度（Multi-Agent Deep Deterministic Policy Gradient，MADDPG）的故障自愈机制，该机制在每个子网部署独立智能体，通过GNN编码器融合局部观测与拓扑结构，使智能体具备全局拓扑感知能力，并利用集中式Critic促进多智能体协作决策，实现网络故障自愈。同时，设计了奖励函数联合优化恢复率、恢复时间与能耗。在Fat‑Tree拓扑上的仿真实验验证了该机制自愈性能和能效优化方面的有效性，以及在不同故障密度、动态流量与非线性功耗模型下的鲁棒性。在真实GÉANT拓扑上的泛化实验验证了该机制的可迁移性。

Abstract

Green computing power network requires fault self-healing while taking into account the recovery speed and energy efficiency. The existing methods mostly adopt centralized control

lack of topology awareness and distributed collaboration capabilities

and less consider energy optimization. Therefore

a fault self-healing mechanism based on Graph Neural Network (GNN) and Multi-Agent Deep Deterministic Policy Gradient (MADDPG) is proposed. The mechanism deploys independent agents in each subnet. The local observation and topology structure are fused by GNN encoder

so that the agent has global topology perception ability. The centralized Critic is used to promote multi-agent collaborative decision-making and realize network fault self-healing. At the same time

a reward function is designed to jointly optimize the recovery rate

recovery time and energy consumption. Simulation experiments on the Fat-Tree topology verify the effectiveness of the self-healing performance and energy efficiency optimization of the mechanism

as well as the robustness under different fault density

dynamic traffic and nonlinear power consumption models. The generalization experiment on the real GÉANT topology verifies the mobility of the mechanism.

关键词

Keywords

references

国家发展改革委 , 国家能源局 , 等 . 数据中心绿色低碳发展专项行动计划 [R ] . 北京 : 国家发展改革委 , 2024 .

National Development and Reform Commission , National Energy Administration , etc . Data Center Green Low-carbon Development Special Action Plan [R ] . Beijing : National Development and Reform Commission , 2024 .

中国信息通信研究院 . 先进计算暨算力发展指数蓝皮书(2025年) [R ] . 北京 : 中国信息通信研究院 , 2026 .

China Academy of Information and Communications Technology . Blue Book of Advanced Computing and Arithmetic Development Index (2025) [R ] . Beijing : China Academy of Information and Communications Technology , 2026 .

谢人超 , 李舒涛 , 胡珉昊 , 等 . 面向绿色计算的算网能一体化: 架构、关键问题与挑战 [J ] . 通信学报 , 2025 , 46 ( 8 ): 205 - 224 .

Xie R C , Li S T , Hu M H , et al . Integration of computing network energy for green computing: architecture, key issues and challenges [J ] . Journal on Communications , 2025 , 46 ( 8 ): 205 - 224 .

Tapolcai J , Babarczi P , Brányi B , et al . Connectivity preserving graph sequences for routing arborescence construction [J ] . IEEE Journal on Selected Areas in Communications , 2025 , 43 ( 2 ): 484 - 494 .

Dou S S , Guo Z H . Path programmability recovery under controller failures for SD-WANs: recent advances and future research challenges [J ] . IEEE Communications Magazine , 2024 , 62 ( 11 ): 100 - 106 .

Wang W X , Dong P , Qiao W X , et al . MultiS-IDRM: An intelligent disaster recovery mechanism for multi-interface server in emergency communication systems [J ] . IEEE Transactions on Vehicular Technology , 2026 , 75 ( 1 ): 1536 - 1550 .

Li R , Cheng Z , Lee P P C , et al . Automated intelligent healing in cloud-scale data centers [C ] // 40th International Symposium on Reliable Distributed Systems (SRDS 2021) . Chicago : IEEE , 2021 : 244 - 253 .

Mwangi A , Navarro-Hilfiker L , Brewka L , et al . A Threshold-Triggered Deep Q-Network-Based Framework for Self-Healing in Autonomic Software-Defined IIoT-Edge Networks [J ] . IEEE Transactions on Network and Service Management , 2026 , 23 : 1297 - 1311 .

Laheri R , Sukumar H K K , Kola C , et al . Self‑healing infrastructure: leveraging reinforcement learning for autonomous cloud recovery and enhanced resilience [J ] . Journal of Information Systems Engineering and Management , 2025 , 10 ( 49 s): 352 - 357 .

Huang W W , Yu H C , Li Y Y , et al . BPDM-GCN: Backup Path Design Method Based on Graph Convolutional Neural Network [J ] . Future Internet , 2025 , 17 ( 5 ): 1 - 20 .

Zhou J , Zheng W , Wang D , et al . A resilient network recovery framework against cascading failures with deep graph learning [J ] . Proceedings of the Institution of Mechanical Engineers , Part O: Journal of Risk and Reliability, 2024 , 238 ( 1 ): 193 - 203 .

Chen Y Y , Gu A Y , Cui L , et al . MTEAL: network routing optimization of SD‑WAN traffic engineering integrating multi‑dimensional QoS metrics [J ] . Journal of Network and Computer Applications , 2025 , 242 : 104272 .

Lowe R , Wu Y , Tamar A , et al . Multi‑agent actor‑critic for mixed cooperative‑competitive environments [C ] // Advances in Neural Information Processing Systems (NIPS 2017) . Long Beach : MIT Press , 2017 : 6379 - 6390 .

Taveras-Cruz A J , Mariano-Hernández D , Jiménez-Matos E , et al . Adaptive protection based on multi-agent systems for AC microgrids: A review [J ] . Applied Energy , 2025 , 377 ( Part D ): 124673 .

Huang Y M , Han F X , Zhao S J . HT-DRIVE Heterogeneous-Temporal GNN-Aided Multi-Agent Reinforcement Learning for Resource Allocation in Platoon-Based V2X Networks [J ] . IEEE Transactions on Communications , 2026 , 74 : 6541 - 6557 .

Gao H J , Jiang S Y , Li Z M , et al . A Two-Stage Multi-Agent Deep Reinforcement Learning Method for Urban Distribution Network Reconfiguration Considering Switch Contribution [J ] . IEEE Transactions on Power Systems , 2024 , 39 ( 6 ): 7064 - 7076 .

Chanda D , Soltani N Y . A Heterogeneous Graph-Based Multi-Task Learning for Fault Event Diagnosis in Smart Grid [J ] . IEEE Transactions on Power Systems , 2025 , 40 ( 2 ): 1427 - 1438 .

LIU C , XU F , GAO C , et al . Deep learning resilience inference for complex networked systems [J ] . Nature Communications , 2024 , 15 ( 1 ): 9203 .

Peng K X , Qi H W , Ma T H . Graph based multi-agent reinforcement learning with evolutionary population for cooperation [J ] . Neural Networks , 2026 , 196 : 108437 .

Liu Z , Zhang J , Shi E , et al . Graph Neural Network Meets Multi-Agent Reinforcement Learning: Fundamentals, Applications, and Future Directions [J ] . IEEE Wireless Communications , 2024 , 31 ( 6 ): 39 - 47 .

Yang Y , Goia F . Hyperparameter optimization impact and tuning guidelines for decentralized multi-agent reinforcement learning in multi-energy neighborhoods [J ] . Applied Energy , 2026 , 414 : 127833 .

Zhao J , Huang K , Gao Y , et al . Coordinated scheduling optimization for Computility center microgrid considering computing resources dynamic pooling [J ] . Applied Energy , 2025 , 393 : 125971 .

Al-Fares M , Loukissas A , Vahdat A . A scalable, commodity data center network architecture [C ] // Proceedings of the ACM SIGCOMM 2008 Conference on Data Communication . New York : ACM , 2008 : 63 - 74 .

罗必雄 , 张力 , 句赫 , 等 . 面向绿色计算的算网能一体化协同任务调度机制研究 [J ] . 通信学报 , 2025 , 46 ( 9 ): 32 - 46 .

LUO B X , ZHANG L , JU H , et al . Computing-network-energy integrated collaborative task scheduling mechanism for green computing [J ] . Journal on Communications , 2025 , 46 ( 9 ): 32 - 46 .

Jin C , Bai X , Yang C , et al . A review of power consumption models of servers in data centers [J ] . Applied Energy , 2020 , 265 : 114806 .

Tröpgen H , Schöne R , Ilsche T , et al . 16 Years of SPEC Power: An Analysis of x86 Energy Efficiency Trends [C ] // 2024 IEEE International Conference on Cluster Computing Workshops (CLUSTER Workshops) . Kobe, Japan : IEEE , 2024 : 76 - 80 .

Sheraz S M , Arfeen A , Haider U . Energy-efficient data center network infrastructure with network switch refresh model [J ] . IEEE Access , 2023 , 11 : 45066 - 45082 .

Nsaif M , Kovásznai G , Malik A , et al . SM-FPLF: Link-state prediction for software-defined DCN power optimization [J ] . IEEE Access , 2024 , 12 : 79496 - 79518 .

Racedo S , Jaumard B , Glatard T , et al . Advances in power consumption model for data centers: Analytical formulas vs. machine learning models [J ] . Future Generation Computer Systems , 2026 , 176 : 108141 .

Rao N S V , Ma C Y T , He F . Game‑theoretic strategies for cyber‑physical infrastructures under component disruptions [J ] . IEEE Transactions on Reliability , 2023 , 72 ( 2 ): 483 - 497 .

Zhou L M , Liu A L , Liu H W , et al . Root Cause Analysis Method Based on Large Language Models with Residual Connection Structures [PP ] . arXiv ( 2026-02-09 ) [ 2026-07-26 ] . arXiv: 2602.08804 .

Zhou Y , Wei F C , Li S X , et al . Data center load modeling through optimal energy consumption characteristics: A path to simultaneously enhance energy efficiency and demand response quality [J ] . Applied Energy , 2025 , 393 : 126095 .

Lin W W , Yu T H , Gao C Z , et al . A hardware-aware CPU power measurement based on the power-exponent function model for cloud servers [J ] . Information Sciences , 2021 , 547 : 1045 - 1065 .

Radovanovic A , Chen B , Talukdar S , et al . Power modeling for effective datacenter planning and compute management [J ] . IEEE Transactions on Smart Grid , 2022 , 13 ( 2 ): 1611 - 1621 .

Liu J J , Zhao H , Wang S Y , et al . Real‑time emission and cost estimation based on unit‑level dynamic carbon emission factor [J ] . Energy Conversion and Economics , 2023 , 4 ( 1 ): 47 - 60 .

Views

下载量

CSCD

Alert me when the article has been cited

提交

Tools

Publicity Resources

Multi-head graph attention driven feature aggregation for large-scale remote-sening constellation mission planning algorithm

Heterogeneous graph neural network algorithm based on topology information enhancement

A decomposable and interpretable cellular network traffic prediction model based on graph convolutional neural network

Node classification of complex network based on enhanced graph neural network and contrastive learning

语义驱动的可视化数据挖掘：方法体系及应用研究

Related Author

Xiao Lixia

Zhou Jiaxi

Wei Zixiang

Dai Ao

Zhou Kangyan

She Yuxuan

DENG Kun

YAO Bowu

Related Institution

School of Electronic Engineering, Xidian University

School of Cyber Science and Engineering, 6G Research Center, Huazhong University of Science and Technology

Provincial Key Laboratory of Multimodal Perceiving and Intelligent Systems, Jiaxing University

College of Artificial Intelligence and Information Engineering, East China University of Technology

College of Information Science and Engineering, Jiaxing University

AI问答

Postal code：100079
Tel：（010）53879206 Email：tmw@bjxintong.com.cn
Technical support is provided by Beijing Founder electronics co., LTD 京ICP备09082226号-64 京公网安备11010602201714号
It is recommended to read the content of this site in Chrome&IE9+. Please switch to extreme mode in browser 360.
Cookies We use cookies to help provide and enhance our service and tailor content. By continuing, you agree to the use of cookies.

⁰