| 引用本文: | 朱志龙,秦文萍,宋子宁,李森良,逯瑞鹏,秦鹏慧.基于两级注意力融合的DFIG并网电力系统暂态稳定评估[J].电力自动化设备,2025,45(9):79-87. |
| ZHU Zhilong,QIN Wenping,SONG Zining,LI Senliang,LU Ruipeng,QIN Penghui.Transient stability assessment of DFIG grid-connected power system based on two-level attention fusion[J].Electric Power Automation Equipment,2025,45(9):79-87. |
|
| 摘要: |
| 针对大规模双馈风电机组(DFIG)并网电力系统暂态稳定评估(TSA)中存在的动态时空特征挖掘不充分和无法兼顾不同尺度数据特征的问题，提出基于两级注意力融合(AF)的TSA方法。为了精确模拟DFIG并网电力系统的暂态稳定动态响应并兼顾仿真准确性和快速性，建立DFIG并网机电-电磁暂态混合仿真模型；为了充分挖掘暂态稳定特征，提出基于概率权重因子的特征融合方法，基于Transformer编码器提出两级AF模型，实现暂态稳定数据时空特征融合以及不同暂态尺度数据特征通道间权重值共享及融合。IEEE 10机39节点系统的仿真结果表明：所提模型的评估效果优于其他常用深度学习算法且具备良好的可解释性。 |
| 关键词: DFIG 暂态稳定评估 Transformer模型 注意力融合 机电-电磁暂态混合仿真 |
| DOI：10.16081/j.epae.202507007 |
| 分类号: |
| 基金项目:国家自然科学基金资助项目(52477115) |
|
| Transient stability assessment of DFIG grid-connected power system based on two-level attention fusion |
|
ZHU Zhilong1, QIN Wenping1, SONG Zining1, LI Senliang2, LU Ruipeng1, QIN Penghui1
|
|
1.Key Laboratory of Cleaner Intelligent Control on Coal & Electricity, Ministry of Education, Shanxi Key Laboratory of Power System Operation and Control, Taiyuan University of Technology, Taiyuan 030024, China;2.China Three Gorges Corporation Shanxi Branch, Taiyuan 030024, China
|
| Abstract: |
| Aiming at the problems of insufficient mining of dynamic spatiotemporal features and inability to take the characteristics of data at different scales into account in the transient stability assessment(TSA) of large-scale doubly fed induction generator(DFIG) grid-connected power system, a TSA method based on two-level attention fusion(AF) is proposed. In order to accurately simulate the transient stability dynamic response of DFIG grid-connected power system and take both simulation accuracy and speed into account, a hybrid electromechanical-electromagnetic transient simulation model of DFIG grid-connected system is established. In order to fully explore the transient stability features, a feature fusion method based on probability weight factor is proposed. A two-level AF model based on Transformer encoder is proposed to achieve the fusion of spatiotemporal features of transient stability data and the sharing and fusion of weight values among feature channels of data with different transient scales. The simulative results of IEEE 10-machine 39-bus system show that the evaluation effect of the proposed model is superior to other commonly used deep learning algorithms and it has good interpretability. |
| Key words: DFIG transient stability assessment Transformer model attention fusion hybrid electromechanical-electromagnetic transient simulation |
