基于多变量样本卷积交互网络的电力系统频率安全性评估  

Frequency Safety Assessment of Power Systems Based on Multivariable-sample Convolution and Interaction Network

作　　者：刘杰[1,2] 石访 宋雪萌[2] 田硕硕 聂礼强 LIU Jie;SHI Fang;SONG Xuemeng;TIAN Shuoshuo;NIE Liqiang(National Joint Engineering Laboratory of Power Grid with Electric Vehicles(Shandong University),Jinan 250061,China;School of Computer Science and Technology,Shandong University,Qingdao 266200,China;School of Electrical Engineering,Shandong University,Jinan 250061,China)

机构地区：[1]电动汽车电网接入技术国家地方联合工程实验室(山东大学),山东省济南市250061 [2]山东大学计算机科学与技术学院,山东省青岛市266200 [3]山东大学电气工程学院,山东省济南市250061

出　　处：《电力系统自动化》2024年第22期160-170,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：现有电力系统暂态频率智能评估方法未充分考虑输入数据的时序特征。因此,文中提出一种基于暂态频率响应曲线智能预测的电力系统频率安全性评估方法。设计了一种多变量样本卷积交互网络,可充分挖掘电力系统量测数据的时序特征,从而提高电力系统暂态频率响应曲线的预测精度;基于所预测的频率响应曲线计算最大频率偏差、最大频率偏差发生时间和准稳态频率等关键指标,并综合评估系统的频率安全性。在频率稳定标准算例上进行仿真测试,结果表明,所提方法与深度学习等经典方法相比,频率响应曲线预测和系统频率安全性评估精度均得到有效提升。The existing intelligent transient frequency assessment methods in power systems do not adequately consider the temporal characteristics of input data.Therefore,a frequency safety assessment method for power systems based on intelligent prediction of transient frequency response curves is proposed.A multivariate-sample convolutional interactive network is designed to fully exploit the temporal characteristics of power system measurement data,thereby improving the prediction accuracy of transient frequency response curves of the power system.Key indicators,such as the maximum frequency deviation,occurrence time of the maximum frequency deviation,and the metastability frequency are calculated based on the predicted frequency response curves,and the frequency safety of the system is comprehensively assessed.Simulation tests are conducted on frequency stability standard cases,and the results show that the proposed method effectively improves the accuracies of frequency response curve prediction and system frequency safety assessment compared with classical methods such as deep learning.

关 键 词：频率安全 深度学习 安全性评估 时序预测模型 卷积交互网络 暂态频率响应 

分 类 号：TM73[电气工程—电力系统及自动化]