| 引用本文: | 吴俊勇,史法顺,李栌苏,赵鹏杰,张若愚.基于MRSE-CNN的电力系统多任务暂态稳定自适应评估[J].电力自动化设备,2025,45(2):167-175. |
| WU Junyong,SHI Fashun,LI Lusu,ZHAO Pengjie,ZHANG Ruoyu.Multi-task transient stability adaptive assessment of power system based on MRSE-CNN[J].Electric Power Automation Equipment,2025,45(2):167-175. |
|
| 摘要: |
| 为解决暂态功角与暂态电压一体化评估中可靠性不足、在线更新耗时过长的问题，提出多任务暂态稳定自适应评估方法。利用变步长二分法从时间维度上构建暂态功角与暂态电压的稳定边界；提出一种融合多尺度残差挤压激励机制的多任务卷积神经网络，该网络直接面向量测数据，仅需3个采样点即可完成输入特征与稳定边界的映射，在保证快速性的基础上实现了高精度的边界拟合；通过引入自适应动态权重的Huber损失函数进一步增强模型在实际应用中的可靠性；在线应用时，通过迁移学习实现模型在负荷、拓扑、新能源3个维度下的自适应更新。在改进的IEEE 39节点系统中的验证结果表明，所提方法不仅兼顾快速性、准确性与可靠性，而且在未知场景下具备快速更新的能力。 |
| 关键词: 暂态稳定评估 暂态功角稳定 暂态电压稳定 极限切除时间 自适应动态权重 迁移学习 |
| DOI：10.16081/j.epae.202411016 |
| 分类号:TM712 |
| 基金项目:国家重点研发计划项目(2018YFB0904500) |
|
| Multi-task transient stability adaptive assessment of power system based on MRSE-CNN |
|
WU Junyong1, SHI Fashun1, LI Lusu1, ZHAO Pengjie2, ZHANG Ruoyu3
|
|
1.School of Electrical Engineering, Beijing Jiaotong University, Beijing 100044, China;2.State Grid Shanxi Electric Power Company, Taiyuan 030021, China;3.Institute of Science and Technology of China Three Gorges Corporation, Beijing 100038, China
|
| Abstract: |
| In order to solve the problems of insufficient reliability and long time-consuming online update in the integrated assessment of transient power angle and transient voltage, a multi-task transient stability adaptive assessment method is proposed. The variable step size dichotomy method is used to construct the stable boundary of transient power angle and transient voltage from the time dimension. A multi-task convolutional neural network incorporating a multi-scale residual squeeze excitation mechanism is proposed, which is directly oriented to the measurement data and requires only three sampling points to complete the mapping between the input features and the stabilized boundaries, and realizes high-precision boundary fitting on the basis of guaranteeing the rapidity. The reliability of the model in practical application is further enhanced by introducing the Huber loss function with adaptive dynamic weight. Adaptive updating of the model under three dimensions of load, topology, and renewable energy is realized by transfer learning when applied online. The validation results in an improved IEEE 39-bus system show that the proposed method not only balances the rapidity, accuracy and reliability, but also has the capability of fast updating under unknown scenarios. |
| Key words: transient stability assessment transient power angle stability transient voltage stability critical clear time adaptive dynamic weight transfer learning |
