基于多尺度图注意力网络的电力系统暂态稳定评估  

Transient Stability Assessment of Power Systems Based on Multi-scale Graph Attention Network

作　　者：傅太国屹 杜友田[1] 吕昊 李宗翰 刘俊[3] FU Taiguoyi;DU Youtian;LYU Hao;LI Zonghan;LIU Jun(School of Automation Science and Engineering,Xi’an Jiaotong University,Xi’an 710049,China;National Key Laboratory of Grid Security(China Electric Power Research Institute),Beijing 100192,China;School of Electrical Engineering,Xi’an Jiaotong University,Xi’an 710049,China)

机构地区：[1]西安交通大学自动化科学与工程学院,陕西省西安市710049 [2]电网安全全国重点实验室(中国电力科学研究院有限公司),北京市100192 [3]西安交通大学电气工程学院,陕西省西安市710049

出　　处：《电力系统自动化》2025年第3期60-70,共11页Automation of Electric Power Systems

基　　金：国家重点研发计划资助项目(2021YFB2400800)。

摘　　要：已有基于图深度学习的暂态稳定评估方法考虑了电网的拓扑结构特征,但对电网拓扑结构图中多尺度子图间的信息传递特性没有进行有效建模,导致判稳模型对电网局部与全局动态耦合关系的捕捉不足,降低了模型在复杂扰动下的判稳精度。因此,提出了一种融合多尺度子图信息传递过程的功角暂态稳定评估方法。首先,提出并构建了一种k阶图注意力网络,以不同尺度的电网拓扑子图作为图深度学习中特征提取的基本单元。然后,通过注意力机制为特征聚合分配自适应权重,以挖掘实际电网中不同细粒度区域之间的特性。最后,通过CEPRI-TAS-173系统验证了所提方法的可行性和有效性。Existing transient stability assessment methods based on graph deep learning consider the topological structure characteristics of power grids.However,the information transmission characteristics among multi-scale subgraphs in the topological structure of power grids are not effectively modeled,resulting in the insufficient capturing of the local and global dynamic coupling relationship of power grids by the stability judgment model,which reduces the stability judgment accuracy of the model under complex perturbations.Therefore,an assessment method for power angle transient stability integrating the information transmission process of multi-scale subgraphs is proposed.Firstly,a k-dimensional graph attention network is proposed and constructed,which regards the different-scale power grid topology subgraphs as the basic unit for feature extraction in graph deep learning.Then,adaptive weights are assigned to the feature aggregation through the attention mechanism to mine the characteristics between different fine-grained regions in the actual power grid.Finally,the feasibility and effectiveness of the proposed method are verified through the CEPRI-TAS-173 system.

关 键 词：暂态稳定评估 深度学习 多尺度子图 特征提取 图注意力网络 

分 类 号：TM7[电气工程—电力系统及自动化]