史法顺, 吴俊勇, 季佳伸, 步雨洛, 李栌苏, 赵鹏杰, 李宝琴. 基于深度学习的电力系统暂态电压与暂态功角稳定一体化超前评估[J]. 电网技术, 2023, 47(2): 741-754. DOI: 10.13335/j.1000-3673.pst.2022.0725
引用本文: 史法顺, 吴俊勇, 季佳伸, 步雨洛, 李栌苏, 赵鹏杰, 李宝琴. 基于深度学习的电力系统暂态电压与暂态功角稳定一体化超前评估[J]. 电网技术, 2023, 47(2): 741-754. DOI: 10.13335/j.1000-3673.pst.2022.0725
SHI Fashun, WU Junyong, JI Jiashen, BU Yuluo, LI Lusu, ZHAO Pengjie, LI Baoqin. Integrated Advance Assessment of Power System Transient Voltage and Transient Angle Stability Based on Deep Learning[J]. Power System Technology, 2023, 47(2): 741-754. DOI: 10.13335/j.1000-3673.pst.2022.0725
Citation: SHI Fashun, WU Junyong, JI Jiashen, BU Yuluo, LI Lusu, ZHAO Pengjie, LI Baoqin. Integrated Advance Assessment of Power System Transient Voltage and Transient Angle Stability Based on Deep Learning[J]. Power System Technology, 2023, 47(2): 741-754. DOI: 10.13335/j.1000-3673.pst.2022.0725

基于深度学习的电力系统暂态电压与暂态功角稳定一体化超前评估

Integrated Advance Assessment of Power System Transient Voltage and Transient Angle Stability Based on Deep Learning

  • 摘要: 暂态电压稳定(transient voltage stability,TVS)与暂态功角稳定(transient angle stability,TAS)是电力系统安全运行的重要基础。随着新型电力系统的建设,电压与功角问题紧密耦合且频发,亟需高精度的一体化超前评估为紧急控制夯实基础。首先根据调研整合了表征功角稳定与电压稳定的综合特征,并根据极限梯度提升(extreme gradient boosting,XGBoost)衡量特征贡献度,根据贡献度生成含差分特征的特征集作为评估模型的输入。其次,提出了融合挤压激励模块的多尺度卷积门控循环单元模型(a multi-scale convolution gated recurrent unit integrated with squeeze excitation,SE-CGRU)。该模型通过挤压激励(squeeze & excitation,SE)模块实现特征通道权重的自适应调整,并利用多尺度卷积融合细节特征与宏观特征,实现了暂态功角与暂态电压的高精度一体化评估。在线评估时无需已知故障切除时间即可给出预测结果并输出系统当前状态下的安全裕度。通过引入带时间约束的损失函数与动态权重训练的方式,在保持现有精度的基础上大大缩减了响应时间,实现超前评估。采用多判据融合策略进一步减少了漏判与误判,提高了模型评估的可靠性。以新英格兰10机39节点系统和国内某区域省级互联系统为例验证分析,结果表明所提方法可实现高精度的功角和电压稳定一体化超前评估。

     

    Abstract: Transient voltage stability (TVS) and transient angle stability (TAS) are the important bases for the safe operation of the power system. With the construction of the new power systems, the problems of TVS and TAS have been closely coupled and occurred frequently, which needs a high precision integrated advanced evaluation urgently to lay a solid foundation for the emergency control. Firstly, according to the investigation, the comprehensive features of the TAS and TVS are integrated, and the feature contribution degree is measured according to the extreme gradient boosting (XGboost), from which the feature set with differential features is generated as the input of the evaluation model. Secondly, a multi-scale convolution gated recurrent unit model integrating the extrusion excitation (SE-CGRU) is proposed. The model realizes the adaptive adjustment of the feature weight channel through the squeeze and excitation (SE) module, and fuses the detail features and the macro features by using the multi-scale convolution to realize the high-precision integrated evaluation of the transient power angle and transient voltage so that the prediction results can be given without knowing the fault clearing time and the safety margin of the system under the current state can be output during the online evaluation. By introducing the loss function with a time constraint and the dynamic weight training, the response time is greatly reduced and the advance evaluation is realized on the basis of maintaining the existing accuracy. The multi-criterion fusion strategy further reduces the missed judgments and misjudgments, and improves the reliability of model evaluation. Taking the New England 10 machine 39 bus system and a regional provincial interconnected system in China as examples to verify and analyze, the results show that the proposed method can achieve the high-precision integrated advance evaluation of the power angle and voltage stability.

     

/

返回文章
返回