Source-Load Separation Method Based on Multi-Agent Reinforcement Learning

HUA Xingyuan; DUAN Sijing; CUI Wenpeng; CHEN Yuzhe; YANG Qingchen; QIAO Nan; REN Ju

doi:10.12263/DZXB.20260338

您当前的位置：

首页 >

文章列表页 >

Source-Load Separation Method Based on Multi-Agent Reinforcement Learning

更新时间：2026-06-15

- Source-Load Separation Method Based on Multi-Agent Reinforcement Learning
- ACTA ELECTRONICA SINICA Pages: 1-13(2026)
- 作者机构：
  
  1.清华大学计算机科学与技术系，北京 100084
  2.北京智芯微电子科技有限公司，北京 102200
  3.中南大学计算机学院，湖南长沙 410083
- 作者简介：
- 基金信息：
  
  Smartchip Research Project of the Chinese Institute of Electronics(2024-02);The National Natural Science Foundation of China(62432004;62402279);The Fundamental and Interdisciplinary Disciplines Breakthrough Plan of the Ministry of Education of China(JYB2025XDXM122);The Postdoctoral Fellowship Program(BX20250375);The China Postdoctoral Science Foundation(2025M771503)
- DOI：10.12263/DZXB.20260338
  CLC： TP39;TM715
- Received：28 April 2026，
  
  Accepted：21 May 2026，
  
  Online First：15 June 2026，
- 稿件说明：
移动端阅览
HUA Xingyuan, DUAN Sijing, CUI Wenpeng, et al. Source-Load Separation Method Based on Multi-Agent Reinforcement Learning[J/OL]. ACTA ELECTRONICA SINICA, 2026, 1-13.
DOI：

HUA Xingyuan, DUAN Sijing, CUI Wenpeng, et al. Source-Load Separation Method Based on Multi-Agent Reinforcement Learning[J/OL]. ACTA ELECTRONICA SINICA, 2026, 1-13. DOI： 10.12263/DZXB.20260338.

摘要

近年来，全球可持续能源发展的需求日益增长，推动了光伏发电系统在居民、商业及工业场景中的应用。为了保障电网运行安全和提升能源调度效率，如何准确预测用户侧的电力负荷变化尤为关键。然而现有方法忽略了实际用户侧接入电表所采集的数据中往往同时包含光伏发电（源）和用电负荷（荷）信息，二者未被有效分离，导致预测结果存在偏差。本文提出基于多智能体强化学习的源荷分离方法SoLED，解决了用户之间复杂时空依赖关系建模能力不足，以及负荷与光伏发电混合数据未分离导致的预测偏差的问题。具体地，本文首先设计了一种特征提取模块，能够捕捉用户负荷短期与长期尺度上的变化规律，从而提升对电网时序特征的表达能力。然后，基于物理世界的电网结构构建仿真电网环境，通过在仿真环境中模拟增强时序特征对空间关系的敏感性。最后，提出了基于多智能体强化学习的源荷分离方法，利用仿真环境生成反馈来训练解耦模型，解决了数据分离困难导致的解耦不准确问题。通过仿真实验和实际系统验证，本文提出的方法在两个城镇电网真实数据集中的不同天气条件下相较现有方法分别提升5.7%~24.6%和6.6%~24.2%的准确率。该实验结果表明，提出的方法能够准确解耦功率，有效地解决了用户之间复杂时空依赖关系建模能力不足，以及负荷与光伏发电混合数据未分离导致的预测偏差的问题。

Abstract

In recent years

the growing demand for global sustainable energy development has accelerated the deployment of photovoltaic (PV) systems across residential

commercial

and industrial sectors. To ensure the safe operation of power grids and improve energy dispatch efficiency

accurately forecasting user-side power load variations has been increasingly critical. However

existing methods often overlook that the data collected from user-side smart meters typically contain both PV generation (source) and power consumption (load) components

which are not effectively separated

leading to biased prediction results. To address this issue

this paper proposes a multi-agent reinforcement learning-based source-load decoupling method

namely SoLED

that mitigates prediction bias caused by the unseparated mixed data of load and PV generation

as well as insufficient modeling of complex spatiotemporal dependencies among users. Specifically

we first design a feature extraction module to capture both short-term and long-term variations in user load

thereby enhancing the representation of spatiotemporal characteristics within the power grid. Then

based on the physical topology of the grid

we construct a simulation environment that models voltage responses under different load and PV power conditions. Finally

leveraging the feedback generated from this simulation

we train a decoupling model within a multi-agent reinforcement learning framework to achieve accurate source-load separation. Experimental results demonstrate that the proposed method improves prediction accuracy by 5.7%~24.6% and 6.6%~24.2% under different weather conditions on two real-world urban power grid datasets. These results confirm the effectiveness of the proposed approach in accurately decoupling user-side mixed power data and enhancing the modeling capability of complex spatiotemporal dependencies

thereby reducing the prediction bias in user-side load forecasting.

关键词

Keywords

references

Detchon R , Van Leeuwen R . Policy: Bring sustainable energy to the developing world [J ] . Nature , 2014 , 508 ( 7496 ): 309 - 311 . DOI: 10.1038/508309a http://dx.doi.org/10.1038/508309a

Batzelis E , Rather Z H , Barton J , et al . Solar integration in the UK and India: technical barriers and future directions [PP/OL ] . V1. arXiv ( 2021-04-01 )[ 2023-06-15 ] . https://doi.org/10.17028/rd.lboro.14453133.v1 https://doi.org/10.17028/rd.lboro.14453133.v1 .

杨峰 , 刘心松 , 邱元杰 , 等 . 电网调度自动化的任务服务模型和调度算法研究 [J ] . 电子学报 , 2002 , 30 ( 12 ): 1811 - 1815 .

Yang Feng , Liu Xinsong , Qiu Yuanjie , et al . Study on task service model and scheduling algorithm in electric power automatization [J ] . Acta Electronica Sinica , 2002 , 30 ( 12 ): 1811 - 1815 . (in Chinese)

于昕妍 , 沈艳霞 , 陈杰 , 等 . 考虑概率区间的微电网短期负荷多目标预测方法 [J ] . 电子学报 , 2017 , 45 ( 4 ): 930 - 936 .

Yu Xinyan , Shen Yanxia , Chen Jie , et al . A multi-objective prediction method for short-term microgrid load considering interval probability [J ] . Acta Electronica Sinica , 2017 , 45 ( 4 ): 930 - 936 . (in Chinese)

王后能 , 殷劲骁 , 廖小兵 , 等 . 微电网并离网切换系统混合樽海鞘优化控制策略 [J ] . 电子学报 , 2025 , 53 ( 11 ): 4091 - 4103 .

Wang Houneng , Yin Jingxiao , Liao Xiaobing , et al . Hybrid salp swarm optimization control strategy for microgrid grid-connected and islanded mode switching system [J ] . Acta Electronica Sinica , 2025 , 53 ( 11 ): 4091 - 4103 . (in Chinese)

Chen X , Qu G N , Tang Y J , et al . Reinforcement learning for selective key applications in power systems: Recent advances and future challenges [J ] . IEEE Transactions on Smart Grid , 2022 , 13 ( 4 ): 2935 - 2958 . DOI: 10.1109/tsg.2022.3154718 http://dx.doi.org/10.1109/tsg.2022.3154718

吴忠强 , 程洪强 . 网络攻击下考虑状态受限的微电网安全运行与控制 [J ] . 电子学报 , 2024 , 52 ( 9 ): 3240 - 3250 . DOI: 10.12263/DZXB.20221092 http://dx.doi.org/10.12263/DZXB.20221092

Wu Zhongqiang , Cheng Hongqiang . Safe operation and control of microgrid considering state constraints under network attacks [J ] . Acta Electronica Sinica , 2024 , 52 ( 9 ): 3240 - 3250 . (in Chinese) . DOI: 10.12263/DZXB.20221092 http://dx.doi.org/10.12263/DZXB.20221092

Xu F Y , Liu S Y , Qing Y P , et al . Temporal prototype-aware learning for active voltage control on power distribution networks [C ] // Proceedings of the 30th ACM SIGKDD Conference on Knowledge Discovery and Data Mining . New York : ACM , 2024 : 3598 - 3609 . DOI: 10.1145/3637528.3671790 http://dx.doi.org/10.1145/3637528.3671790

Liu S Y , Luo W , Zhou Y Z , et al . Transmission interface power flow adjustment: A deep reinforcement learning approach based on multi-task attribution map [J ] . IEEE Transactions on Power Systems , 2024 , 39 ( 2 ): 3324 - 3335 . DOI: 10.1109/tpwrs.2023.3298007 http://dx.doi.org/10.1109/tpwrs.2023.3298007

Shi X J , Chen Z R , Wang H , et al . Convolutional LSTM Network: A machine learning approach for precipitation nowcasting [C ] // Proceedings of the 29th International Conference on Neural Information Processing Systems - Volume 1 . New York : ACM , 2015 : 802 - 810 .

Vaswani A , Shazeer N , Parmar N , et al . Attention is all you need [C ] // Proceedings of the 31st International Conference on Neural Information Processing Systems . New York : ACM , 2017 : 6000 - 6010 .

Wu Z H , Pan S R , Chen F W , et al . A comprehensive survey on graph neural networks [J ] . IEEE Transactions on Neural Networks and Learning Systems , 2021 , 32 ( 1 ): 4 - 24 . DOI: 10.1109/tnnls.2020.2978386 http://dx.doi.org/10.1109/tnnls.2020.2978386

Masters C L . Voltage rise: The big issue when connecting embedded generation to long 11 kV overhead lines [J ] . Power Engineering Journal , 2002 , 16 ( 1 ): 5 - 12 . DOI: 10.1049/pe:20020101 http://dx.doi.org/10.1049/pe:20020101

Yang G Y , Marra F , Juamperez M , et al . Voltage rise mitigation for solar PV integration at LV grids: Studies from PVNET. dk [J ] . Journal of Modern Power Systems and Clean Energy , 2015 , 3 ( 3 ): 411 - 421 . DOI: 10.1007/s40565-015-0132-0 http://dx.doi.org/10.1007/s40565-015-0132-0

Agalgaonkar Y P , Pal B C , Jabr R A . Distribution voltage control considering the impact of PV generation on tap changers and autonomous regulators [J ] . IEEE Transactions on Power Systems , 2014 , 29 ( 1 ): 182 - 192 . DOI: 10.1109/tpwrs.2013.2279721 http://dx.doi.org/10.1109/tpwrs.2013.2279721

Huang J , Wei Q , Wang T K , et al . Spatio-temporal feature fusion-based hybrid GAT-CNN-LSTM model for enhanced short-term power load forecasting [J ] . Energies , 2025 , 18 ( 21 ): 5686 . DOI: 10.3390/en18215686 http://dx.doi.org/10.3390/en18215686

Su Z E , Zheng G Q , Hu M S , et al . Short-term load forecasting of regional integrated energy system based on spatio-temporal convolutional graph neural network [J ] . Electric Power Systems Research , 2024 , 232 : 110427 . DOI: 10.1016/j.epsr.2024.110427 http://dx.doi.org/10.1016/j.epsr.2024.110427

Cheng M , Yu J W , Wu M K , et al . A spatiotemporal deep learning framework for joint load and renewable energy forecasting in stability-constrained power systems [J ] . Information , 2025 , 16 ( 8 ): 662 . DOI: 10.3390/info16080662 http://dx.doi.org/10.3390/info16080662

Chung J , Gulcehre C , Cho K , et al . Empirical evaluation of gated recurrent neural networks on sequence modeling [PP/OL ] . V1.arXiv ( 2014-12-11 )[ 2023-06-15 ] . https://doi.org/10.48550/arXiv.1412.3555 https://doi.org/10.48550/arXiv.1412.3555 .

LeCun Y , Bottou L , Bengio Y , et al . Gradient-based learning applied to document recognition [J ] . Proceedings of the IEEE , 1998 , 86 ( 11 ): 2278 - 2324 . DOI: 10.1109/5.726791 http://dx.doi.org/10.1109/5.726791

Liu Y C , Zhang J H , Fang L J , et al . Multimodal motion prediction with stacked transformers [C ] // 2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition . Piscataway : IEEE , 2021 : 7573 - 7582 . DOI: 10.1109/cvpr46437.2021.00749 http://dx.doi.org/10.1109/cvpr46437.2021.00749

Kipf T N , Welling M . Semi-Supervised Classification with Graph Convolutional Networks [C ] // International Conference on Learning Representation . Washington DC : ICLR , 2017 .

Javed F , Arshad N , Wallin F , et al . Forecasting for demand response in smart grids: An analysis on use of anthropologic and structural data and short term multiple loads forecasting [J ] . Applied Energy , 2012 , 96 : 150 - 160 . DOI: 10.1016/j.apenergy.2012.02.027 http://dx.doi.org/10.1016/j.apenergy.2012.02.027

Yuan Y , Yuan X P , Wang H Y , et al . Net load forecasting method in distribution grid planning based on LSTM network [J ] . Science and Technology for Energy Transition , 2024 , 79 : 57 . DOI: 10.2516/stet/2024062 http://dx.doi.org/10.2516/stet/2024062

Kumar N . Optimized power distribution using machine learning for load forecasting, fault detection, and voltage regulation [J ] . International Journal of Research and Review in Applied Science, Humanities, and Technology , 2024 : 94 - 102 . DOI: 10.71143/0m5nm797 http://dx.doi.org/10.71143/0m5nm797

Awad M , Khanna R . Support vector regression [M ] // Efficient Learning Machines: Theories, Concepts, and Applications for Engineers and System Designers . Berkeley, CA : Apress , 2015 : 67 - 80 . DOI: 10.1007/978-1-4302-5990-9_4 http://dx.doi.org/10.1007/978-1-4302-5990-9_4

Wang G , Xie X Y , Zhang F , et al . bCharge: Data-driven real-time charging scheduling for large-scale electric bus fleets [C ] // 2018 IEEE Real-Time Systems Symposium . Piscataway : IEEE , 2018 : 45 - 55 . DOI: 10.1109/rtss.2018.00015 http://dx.doi.org/10.1109/rtss.2018.00015

Zaremba W , Sutskever I , Vinyals O . Recurrent neural network regularization [PP/OL ] . V5. arXiv ( 2015-02-19 )[ 2023-06-15 ] . https://doi.org/10.48550/arXiv.1409.2329 https://doi.org/10.48550/arXiv.1409.2329 .

Zhou H Y , Zhang S H , Peng J Q , et al . Informer: Beyond efficient transformer for long sequence time-series forecasting [J ] . Proceedings of the AAAI Conference on Artificial Intelligence , 2021 , 35 ( 12 ): 11106 - 11115 . DOI: 10.1609/aaai.v35i12.17325 http://dx.doi.org/10.1609/aaai.v35i12.17325

Lu S , Bao T T . Short-term electricity load forecasting based on NeuralProphet and CNN-LSTM [J ] . IEEE Access , 2024 , 12 : 76870 - 76879 . DOI: 10.1109/access.2024.3407094 http://dx.doi.org/10.1109/access.2024.3407094

Jia Y L , Wang J , Hosseini M R , et al . Temporal graph attention network for building thermal load prediction [J ] . Energy and Buildings , 2024 , 321 : 113507 . DOI: 10.1016/j.enbuild.2023.113507 http://dx.doi.org/10.1016/j.enbuild.2023.113507

Orhi U , Guven C , Stowell D . Enhanced load forecasting with GAT-LSTM: leveraging grid and temporal features [PP/OL ] . V1. arXiv ( 2025-02-12 )[ 2023-06-15 ] . https://doi.org/10.48550/arXiv.2502. 08376 https://doi.org/10.48550/arXiv.2502.08376 .

Yarats D , Fergus R , Lazaric A , et al . Reinforcement learning with prototypical representations [C ] // Proceedings of the 38th International Conference on Machine Learning . New York : PMLR , 2021 : 11920 - 11931 .

Zhang Z X , Liu Q , Wang H , et al . ProtGNN: Towards self-explaining graph neural networks [J ] . Proceedings of the AAAI Conference on Artificial Intelligence , 2022 , 36 ( 8 ): 9127 - 9135 . DOI: 10.1609/aaai.v36i8.20898 http://dx.doi.org/10.1609/aaai.v36i8.20898

Mnih V , Kavukcuoglu K , Silver D , et al . Human-level control through deep reinforcement learning [J ] . Nature , 2015 , 518 ( 7540 ): 529 - 533 . DOI: 10.1038/nature14236 http://dx.doi.org/10.1038/nature14236

Schulman J , Moritz P , Levine S , et al . High-dimensional continuous control using generalized advantage estimation [PP/OL ] . V6. arXiv ( 2018-10-20 )[ 2023-06-15 ] . https://doi.org/10.48550/arXiv.1506.02438 https://doi.org/10.48550/arXiv.1506.02438 .

Fujimoto S , Van Hoof H , Meger D . Addressing function approximation error in actor-critic methods [C ] // Proceedings of the 35th International Conference on Machine Learning . New York : PMLR , 2018 : 1587 - 1596 . DOI: 10.48550/arXiv.1802.09477 http://dx.doi.org/10.48550/arXiv.1802.09477

Schulman J , Wolski F , Dhariwal P , et al . Proximal policy optimization algorithms [PP/OL ] . V2. arXiv ( 2017-08-28 )[ 2023-06-15 ] . https://doi.org/10.48550/arXiv.1707.06347 https://doi.org/10.48550/arXiv.1707.06347 .

Bilal M , Lopez L C . Temporal Graph MLP Mixer for Spatio-Temporal Forecasting [PP/OL ] . V1. arXiv ( 2025-01-17 )[ 2023-06-15 ] . https://doi.org/10.48550/arXiv.2501.10214 https://doi.org/10.48550/arXiv.2501.10214 .

Wang X , Wang X J , Huang F L , et al . DSTF: A Diversified Spatio-Temporal Feature Extraction Model for traffic flow prediction [J ] . Neurocomputing , 2025 , 621 : 129280 . DOI: 10.1016/j.neucom.2024.129280 http://dx.doi.org/10.1016/j.neucom.2024.129280

Gupta N , Hare J Z , Kannan R , et al . Deep Meta Coordination Graphs for Multi-agent Reinforcement Learning [PP/OL ] . V1. arXiv ( 2026-02-10 )[ 2023-06-15 ] . https://doi.org/10.48550/arXiv.2502.04028 https://doi.org/10.48550/arXiv.2502.04028 .

Wu J , Huang Z X . Promoting diversity in mixed complex cooperative and competitive multi-agent environment [C ] // Proceedings of the 32nd ACM International Conference on Information and Knowledge Management . New York : ACM , 2023 : 4355 - 4359 . DOI: 10.1145/3583780.3615217 http://dx.doi.org/10.1145/3583780.3615217

Bayen A , Gao J X , Velu A , et al . The surprising effectiveness of PPO in cooperative multi-agent games [C ] // Advances in Neural Information Processing Systems 35 . Neural Information Processing Systems Foundation, Inc. (NeurIPS) , 2022 : 24611 - 24624 . DOI: 10.52202/068431-1787 http://dx.doi.org/10.52202/068431-1787

Views

下载量

CSCD

Alert me when the article has been cited

提交

Tools

Publicity Resources

LLM-Enhanced Maritime UAV Routing Algorithm Against Gray-Hole Attacks

Multi-Agent Reinforcement Learning: From Foundational Theory to Cutting-Edge Algorithms

Multimodal Smart Contract Vulnerability Detection Enhanced by Large Language Model Semantics

Image Classification Network with Attenuation Disentangling Mechanism

Low Sidelobe and High Aperture Efficiency Reflectarray Based on Metasurface

Related Author

LI Junwei

RUAN Shulan

LIANG Jiaxuan

LIU Yu

HE You

WU Nan

WEI Shiyu

ZHANG Tingting

Related Institution

Shenzhen International Graduate School, Tsinghua University

School of Computer Science and Technology, Harbin Institute of Technology

Department of Electronic Engineering, Tsinghua University

School of Information and Electronics, Beijing Institute of Technology

School of Integrated Circuits and Electronics, Beijing Institute of Technology

⁰