基于强化学习的电网数据中台异常判别与分类方法

doi:10.3969/j.issn.1008-0198.2026.03.016

湖南电力 ›› 2026, Vol. 46 ›› Issue (3): 121-129.doi: 10.3969/j.issn.1008-0198.2026.03.016

• 电力人工智能与数字化 • 上一篇下一篇

基于强化学习的电网数据中台异常判别与分类方法

蒋广^1,2, 薛静远^1,2, 胡翔^1,2, 曹杰^1,2, 方彬^1,2

1.国网湖南电力有限公司信息通信分公司,湖南长沙 410004;
2.泛在电力物联网湖南省重点实验室,湖南长沙 410004

收稿日期:2025-11-26 修回日期:2025-12-17 出版日期:2026-06-25 发布日期:2026-07-07
通信作者: 蒋广(1996),男,硕士,研究方向为电力大数据应用。
基金资助:
国网湖南省电力有限公司科技项目（5216A623000W）

Anomaly Detection and Classification Methods for Power Grid Data Middle Platform Based on Reinforcement Learning

JIANG Guang^1,2, XUE Jingyuan^1,2, HU Xiang^1,2, CAO Jie^1,2, FANG Bin^1,2

1. Information and Communication Branch of State Grid Hunan Electric Power Company Limited, Changsha 410004, China;
2. Hunan Provincial Key Laboratory of Internet of Things in Electricity, Changsha 410004, China

Received:2025-11-26 Revised:2025-12-17 Online:2026-06-25 Published:2026-07-07

摘要/Abstract

摘要： 为提升电网数据中台运维处置效率,提出一种融合近端策略优化与多头自注意力机制的强化学习模型。模型对电网数据中台异常任务进行实时判别与分类,辅助现场运维人员快速定位异常原因,并针对性地进行异常处置。首先,通过门控循环单元网络实现时间序列依赖关系建模。其次,引入多头自注意力层来精准捕获局部显著性特征。随后,针对电网数据中台异常判别场景特点,设计混合激励函数,确保模型在判别已知异常分类和自主发现新异常分类方面均有较好性能。最后,依托某网省公司的数据中台实际运行数据进行效果验证。实验结果表明,模型总体异常分类准确率达到 98.2%,针对未知新异常类型判别的 F1 Score达到91.5%,显著优于其他基础模型。

关键词: 电网数据中台, 异常判别, 强化学习, 近端策略优化, 注意力机制

Abstract: To improve the efficiency of power grid data platform operation and maintenance, this paper proposes a reinforcement learning model that integrates proximal policy optimization and multi-head self-attention mechanism. This model aims to perform real-time identification and classification of anomalies in the power grid data platform, assisting on-site maintenance personnel in quickly locating the causes of anomalies and carrying out targeted anomaly handling. First, a gated recurrent unit network is used to model time-series dependencies. Second, a multi-head self-attention layer is introduced to accurately capture local saliency features. Then, considering the characteristics of anomaly identification scenarios in power grid data platforms, a hybrid activation function is designed to ensure good performance in both identifying known anomalies and autonomously discovering new anomaly types. Finally, the model's effectiveness is verified using actual operational data from a provincial power grid company's data middle platform. Experimental results show that the overall anomaly classification accuracy reaches 98.2%, and the F1 score for identifying unknown new anomaly types reaches 91.5%, significantly outperforming other comparative models.

Key words: power grid data middle platform, anomaly detection, reinforcement learning, proximal policy optimization, attention mechanism

中图分类号:

TM727
TP18

蒋广, 薛静远, 胡翔, 曹杰, 方彬. 基于强化学习的电网数据中台异常判别与分类方法[J]. 湖南电力, 2026, 46(3): 121-129.

JIANG Guang, XUE Jingyuan, HU Xiang, CAO Jie, FANG Bin. Anomaly Detection and Classification Methods for Power Grid Data Middle Platform Based on Reinforcement Learning[J]. Hunan Electric Power, 2026, 46(3): 121-129.

参考文献

[1] 郑金辉. 数字化转型背景下的数据中台建设探讨[J]. 数字通信世界,2021(11):56-57.
[2] 李志,费晓璐,郭振. 基于数据中台的电力企业数据资产管理方法研究[J]. 电力信息与通信技术,2020,18(7):76-81.
[3] DUAN J.Deep learning anomaly detection in AI-powered intelligent power distribution systems[J]. Frontiers in Energy Research,2024,12:1364456.
[4] LI X,ZHU Z,ZHANG C,et al.Power data analysis and mining technology in smart grid[J]. Energy Informatics,2024,7(1):93-113.
[5] WANG C,XING S,GAO R,et al.Disentangled dynamic deviation transformer networks for multivariate time series anomaly detection[J]. Sensors,2023,23(3):1104.
[6] YI S,ZHENG S,YANG S,et al.Anomaly detection for asynchronous multivariate time series of nuclear power plants using a temporal-spatial transformer[J]. Sensors,2024,24(9):2845-2865.
[7] LU T,WANG L,ZHAO X.Review of anomaly detection algorithms for data streams[J]. Applied Sciences,2023,13(10):6353.
[8] LIU X,GU L,GONG X,et al.A power data anomaly detection model based on deep learning with adaptive feature fusion[J]. Computers,Materials and Continua,2024,79(3):4045-4061.
[9] GUATO BURGOS M F,MORATO J,VIZCAINO IMACANA F P.A review of smart grid anomaly detection approaches pertaining to artificial intelligence[J]. Applied Sciences,2024,14(3):1194.
[10] QIU J,SHI H,HU Y,et al.Unraveling false positives in unsupervised defect detection models:a study on anomaly-free training datasets[J]. Sensors,2023,23(23):9360.
[11] GUPTA P,TRIPATHY P.Unsupervised learning for real-time data anomaly detection: a comprehensive approach[J]. International Journal of Computer Science and Engineering,2024,11(10):1-11.
[12] MAO W,WANG G,KOU L,et al.Deep domain-adversarial anomaly detection with one-class transfer learning[J]. IEEE/CAA Journal of Automatica Sinica,2023,10(2):524-546.
[13] LUO X,JIANG Y,WANG E,et al.Anomaly detection by using a combination of generative adversarial networks and convolutional autoencoders[J]. EURASIP Journal on Advances in Signal Processing,2022,2022(1):112.
[14] TORABI H,MIRTAHERI S L,GRECO S.Practical autoencoder based anomaly detection by using vector reconstruction error[J]. Cybersecurity,2023,6(1):1.
[15] WANG C,XING S,GAO R,et al.Disentangled dynamic deviation transformer networks for multivariate time series anomaly detection[J]. Sensors,2023,23(3):1104.
[16] SUN S,LIU C,ZHU Y,et al.Deep reinforcement learning for the detection of abnormal data in smart meters[J]. Sensors,2022,22(21):8543.
[17] XING Y,TAN J,ZHANG R,et al.Robust anomaly detection of multivariate time series data via adversarial graph attention BiGRU[J]. Big Data and Cognitive Computing,2025,9(5):122.
[18] KANG H,KANG P.Transformer-based multivariate time series anomaly detection using inter-variable attention mechanism[J]. Knowledge-Based Systems,2024,290:111507.
[19] LIU C,WEI Z,ZHOU L,et al.Multidimensional time series classification with multiple attention mechanism[J].Complex & Intelligent Systems,2024,11(1):14.
[20] ZHOU K,WANG W,HU T,et al.Application of improved asynchronous advantage actor critic reinforcement learning model on anomaly detection[J]. Entropy,2021,23(3):274.
[21] WANG Q,MA Y,ZHAO K,et al.A comprehensive survey of loss functions in machine learning[J]. Annals of Data Science,2022,9(2):187-212.

基于强化学习的电网数据中台异常判别与分类方法

Anomaly Detection and Classification Methods for Power Grid Data Middle Platform Based on Reinforcement Learning

PDF (PC)

可视化

摘要/Abstract

引用本文

使用本文

参考文献

相关文章 6

编辑推荐

Metrics

本文评价

[1]	李逸凡, 张晓东, 张鸿伟, 陈颉. 融合IGDT鲁棒-机会型决策的DQN双策略电动汽车移动充电站协同优化调度[J]. 湖南电力, 2025, 45(6): 9-18.
[2]	朱光明, 胡锦豪, 徐松, 万代. 基于双向长短期记忆网络与注意力机制的锂电池剩余使用寿命预测模型[J]. 湖南电力, 2025, 45(5): 1-7.
[3]	王生祺, 陈海波, 叶金翔. 基于深度学习与局部感知注意力的避雷器套管红外图像分割方法[J]. 湖南电力, 2025, 45(2): 122-128.
[4]	于惠钧, 夏梦, 陈刚, 谭福元, 徐银凤. 基于牛顿-拉弗森优化算法与注意力机制优化TCN-GRU的短期电力负荷预测[J]. 湖南电力, 2024, 44(6): 120-127.
[5]	贺庆辰, 秦斌. 基于改进SAC算法的城轨列车混合储能系统动态功率分配策略[J]. 湖南电力, 2024, 44(4): 11-19.
[6]	王淑强, 刘世件, 黄超, 梁家豪. 火电多市场联合运行策略研究[J]. 湖南电力, 2022, 42(6): 76-82.