[1]张义虎,李恒山,李会骞,等.基于动态设备关联GAT的电力设备状态评估方法[J].机械与电子,2026,44(09):96-102.
　ZHANG Yihu,LI Hengshan,LI Huiqian,et al.A Dynamic Association Graph Attention Network for Power Equipment Condition Assessment[J].Machinery & Electronics,2026,44(09):96-102.
点击复制

基于动态设备关联GAT的电力设备状态评估方法()
分享到：

《机械与电子》[ISSN:1001-2257/CN:52-1052/TH]

卷:
44
期数:
2026年09期
页码:
96-102
栏目:
电力控制
出版日期:
2026-09-25

文章信息/Info

Title:
A Dynamic Association Graph Attention Network for Power Equipment Condition Assessment
文章编号:
1001-2257(2026)09-0096-07
作者:
张义虎1; 李恒山1; 李会骞2; 段旗城1; 刘 颖1
1.国能(甘肃)新能源有限公司,甘肃 白银 730900;
2.国家能源集团甘肃电力有限公司,甘肃 兰州 730030
Author(s):
ZHANG Yihu1; LI Hengshan1; LI Huiqian2; DUAN Qicheng1; LIU Ying1
(1.Guoneng (Gansu) New Energy Co.,Ltd.,Baiyin 730900,China;
2.Gansu Electric Power Co.,Ltd.,National Energy Group,Lanzhou 730030,China)
关键词:
电力变压器; 图注意力网络; 多源信息融合; 动态关联; 状态评估
Keywords:
power transformer; graph attention network; multi-source information fusion; dynamic association; condition assessment
分类号:
TM507
文献标志码:
A
摘要:
针对传统电力变压器状态评估方法依赖单一信息源、难以刻画设备间关联耦合的问题,提出一种基于动态设备关联图注意力网络(DAGAT)的多源信息融合状态评估方法。该方法融合电气试验特征、油化试验特征与溶解气体分析(DGA)特征构建多维节点表示,综合电气连接、负载相似度与空间距离定义设备间的复合关联边;进而通过多层图注意力网络编码器实现设备间信息交互,并引入时序注意力机制实现关联权重的动态更新。在电力变压器多源融合数据集上的实验表明,所提方法准确率达到96.8%,宏平均F1值为96.2%,较标准图注意力网络提升2.6百分点。消融实验、噪声鲁棒性实验与小样本实验进一步验证了各模块的有效性与模型的泛化能力。
Abstract:
To address the limitations of the conventional power transformer condition assessment methods that rely on a single information source and fail to capture the coupling relationships among equipment units,this paper proposes a multi-source information fusion condition assessment method based on a Dynamic Association Graph Attention Network (DAGAT).The proposed method constructs multi dimensional node representations by integrating electrical test features,oil chemistry test features,and dissolved gas analysis (DGA) features,and defines composite association edges among equipment units by jointly considering electrical connectivity,load similarity,and spatial proximity.Furthermore,a multilayer graph attention network encoder is used to enable information interaction among equipment units,while a temporal attention mechanism is introduced to enable dynamic updates of the association weights.Experimental results on a multi-source fusion dataset of power transformers demonstrate that the proposed method achieves an accuracy of 96.8% and a macro average F1 score of 96.2%,outperforming the standard graph attention network by 2.6 percentage points.Ablation studies,noise robustness experiments,and few-shot experiments further validate the effectiveness of each module and the generalization capability of the proposed model.

参考文献/References:

[1] 马桂琴,李卫疆,曹洪建,等.电力设备状态监测与智能预警技术的应用研究[J].信息与电脑,2026,38(9):127-129.

[2] 齐子豪,仝杰,张中浩,等.基于多粒度知识特征和Transformer网络的电力变压器故障声纹辨识方法[J].中国电机工程学报,2025,45(4):1311-1322.
[3] 赵文清,祝玲玉,高树国,等.基于多源信息融合的电力变压器故障诊断方法研究[J].电力信息与通信技术,2018,16(10):25-30.
[4] 康佳宇,张沈习,张庆平,等.基于ANOVA 和BO SVM 的变压器故障诊断方法[J].高电压技术,2023,49(5):1882-1891.
[5] Jina E,Zhang Yunyi,He Wei,et al.A new method for oil-immersed transformers fault diagnosis based on evidential reasoning rule with optimized probabilistic distributed[J].IEEE Access,2024,12:34289-34305.
[6] 韦强宇,吐松江·卡日,张紫薇,等.小样本条件下基于改进原型网络的电力变压器故障诊断方法[J].电子测量技术,2026,49(13):152-162.
[7] Wang Lin,Littler T,Liu Xueqin.Dynamic incipient fault forecasting for power transformers using an LSTM model[J].IEEE Transactions on Dielectrics and Electrical Insulation,2023,30(2):563-572.
[8] Li Zihao,He Yigang,Xing Zhikai,et al.Transformer fault diagnosis based on improved deep coupled dense convolutional neural network[J].Electric Power Systems Research,2022,209:107969.
[9] Taha I B M,Ibrahim S,Mansour DEA.Power transformer fault diagnosis based on DGA using a convolutional neural network with noise in measurements[J].IEEE Access,2021,9:111162-111170.
[10] 陈勇,李松,晋伟平,等.基于体素注意力网络的电力设备目标检测模型[J].浙江电力,2024,43(4):121-128.
[11] 钟智,张逸,林湘宁,等.基于图注意力深度网络的电力系统暂态稳定评估[J].电网技术,2021,45(6):2122-2130.
[12] Huang Jiyu,Guan Lin,Su Yinsheng,et al.A topology adaptive high-speed transient stability assessment scheme based on multi-graph attention network with residual structure[J].International Journal of Electrical Power and Energy Systems,2021,130:106984.
[13] Xing Zhikai,He Yigang.Multi-modal information analysis for fault diagnosis with time series data from power transformer[J].International Journal of Electrical Power and Energy Systems,2023,144:108567.
[14] Huang Yuanfang,Huang Zhanhong,Chen Junbin.Multi source data fusion and multi-task physics informed Transformer for power transformer fault diagnosis[J].Energies,2026,19(3):599.

相似文献/References:

[1]王 峰.基于模糊神经网络的电力变压器绝缘状态预测方法[J].机械与电子,2022,(03):50.
　WANG Feng.Prediction Method of Insulation State of Power Transformer Based on Fuzzy Neural Network[J].Machinery & Electronics,2022,(09):50.
[2]王子伟,疏佳铭,陆 斌,等.基于 GASF 及双输入 AlexNet-SVM 的变压器励磁涌流检测[J].机械与电子,2025,(08):30.
　WANG Ziwei,SHU Jiaming,LU Bin,et al.Transformer Inrush Current Detection Based on GASF and Dual-input AlexNet-SVM[J].Machinery & Electronics,2025,(09):30.
[3]郭 毅,车艳亮,伏学文,等.基于退化感知多尺度Transformer的电力设备状态评估方法[J].机械与电子,2026,44(09):120.
　GUO Yi,CHE Yanliang,FU Xuewen,et al.Degradation-aware Multi-scale Transformer for Power Equipment Condition Assessment[J].Machinery & Electronics,2026,44(09):120.

备注/Memo

备注/Memo:
收稿日期:2026-05-25
基金项目:国能(甘肃)新能源有限公司科技项目(E586700018)
作者简介:张义虎 (1986-),男,甘肃秦安人,硕士,研究方向为电力设备在线监测。
更新日期/Last Update: 2026-10-13