| 引用本文: | 周芷汀,李辉,郑杰,谭宏涛,张振,张浩,向学位.物理信息与时空图数据联合驱动的海上风电并网系统频率响应预测[J].电力自动化设备,2025,45(12):82-90. |
| ZHOU Zhiting,LI Hui,ZHENG Jie,TAN Hongtao,ZHANG Zhen,ZHANG Hao,XIANG Xuewei.Frequency response prediction of offshore wind-integrated power system driven by physics-informed spatiotemporal graph learning[J].Electric Power Automation Equipment,2025,45(12):82-90. |
|
| 摘要: |
| 海上风电大规模并网改变了电力系统频率动态特性，准确预测扰动后的频率响应对系统安全稳定运行至关重要。现有频率响应预测方法未充分考虑电网拓扑结构和风电调频机理，导致时空特征利用不足且缺乏可解释性。为此，提出一种物理信息与时空图数据联合驱动的海上风电并网系统频率响应预测方法。建立计及虚拟惯量综合调频控制策略的风电机组动态模型，在此基础上构建风电并网系统频率响应解析模型；设计电力系统时空图卷积网络架构，通过图卷积层挖掘节点间的空间拓扑关系，利用时间卷积层提取频率相关时序特征；采用物理信息嵌入方法将风电并网系统解析频率响应值作为增强特征融入网络架构；在改进的IEEE 39节点和IEEE 14节点系统上进行验证。结果表明所提方法相比传统深度学习方法对频率响应曲线和关键频率指标的预测精度提升，且在不同风电渗透率下表现出良好的泛化能力。 |
| 关键词: 海上风电并网系统 频率响应预测 时空图卷积网络 虚拟惯量综合调频控制 物理信息嵌入 |
| DOI：10.16081/j.epae.202510019 |
| 分类号: |
| 基金项目:重庆市科技创新重大研发项目(CSTB2024TIAD-STX0019) |
|
| Frequency response prediction of offshore wind-integrated power system driven by physics-informed spatiotemporal graph learning |
|
ZHOU Zhiting1, LI Hui1, ZHENG Jie1,2, TAN Hongtao3, ZHANG Zhen1, ZHANG Hao1, XIANG Xuewei1
|
|
1.State Key Laboratory of Power Transmission Equipment Technology, Chongqing University, Chongqing 400044, China;2.CSSC(Chongqing) Haizhuang Wind Power Co.,Ltd.,Chongqing 400044, China;3.Dongfang Electric Autocontrol Engineering Co.,Ltd.,Deyang 618030, China
|
| Abstract: |
| Large-scale offshore wind-integrated has changed the frequency dynamics characteristic of power system. Accurate prediction of frequency response after disturbances is essential for maintaining system safety and stability. Current frequency response prediction methods do not fully consider grid topology and wind power frequency regulation mechanisms, leading to insufficient spatiotemporal feature utilization and a lack of interpretability. To address this issue, a physics-informed spatiotemporal graph learning method of frequency response prediction for offshore wind-integrated power system is proposed. A dynamic model of wind turbine generators with combined virtual inertia modulation control strategy is established, and an analytical frequency response model for wind-integrated power system is constructed based on this model. A structure of spatiotemporal graph convolutional neural network for power system is designed, where graph convolutional layers are utilized to capture spatial topological relationships, temporal convolutional layers are utilized to extract time-related frequency features. A physics-informed embedding approach is employed to integrate analytical frequency response values as enhanced features into the network. The validation is carried out on the modified IEEE 39-bus and IEEE 14-bus systems, the results demonstrate that compared with the conventional deep learning methods, the proposed method improves the prediction accuracy of frequency response curve and key frequency indices, and shows strong generalization ability under different wind power penetration levels. |
| Key words: offshore wind-integrated power system frequency response prediction spatiotemporal graph convolutional network combined virtual inertia modulation control physics-informed embedding |
