数据驱动的风电场全工况非线性调频动态建模  

Data-driven Dynamic Modeling of Nonlinear Frequency Regulation for Wind Farm Under All Operation Conditions

作　　者：胡阳[1,2] 姚欣然[2] 房方 宋子秋 郭强[3] 刘吉臻[1,2] HU Yang;YAO Xinran;FANG Fang;SONG Ziqiu;GUO Qiang;LIU Jizhen(Key Laboratory of Power Station Energy Transfer Conversion and System,Ministry of Education(North China Electric Power University),Beijing 102206,China;School of Control and Computer Engineering,North China Electric Power University,Beijing 102206,China;Electric Power Research Institute of State Grid Shanxi Electric Power Company,Taiyuan 030001,China)

机构地区：[1]电站能量传递转化与系统教育部重点实验室(华北电力大学),北京市102206 [2]华北电力大学控制与计算机工程学院,北京市102206 [3]国网山西省电力公司电力科学研究院,山西省太原市030001

出　　处：《电力系统自动化》2023年第15期162-169,共8页Automation of Electric Power Systems

基　　金：国家重点研发计划资助项目(2021YFE0102400)。

摘　　要：当前风电场级并网调频面临较大挑战。为降低建模复杂度,采用风电场并网点实际运行数据,提出了风电场出力范围内全工况一次调频特性时域差分动态建模方法。首先,进行机理分析并确定输入输出变量,完成数据采样和处理。然后,提出离散工况抽样和非线性评估方法,建立了单工况子空间模型,并进行间隙测度评估全工况调频响应动态非线性度。此后,为高效逼近全工况复杂非线性调频特性,构建混杂有限差分回归向量,并对其张成的差分动态空间进行凸划分,在各子工作域均匀选取建模样本并采用带注意力机制的长短期记忆神经网络进行全工况建模。最后,经过某风电场并网点调频运行数据进行验证,表明所提出的全工况有限差分动态建模方法具有任意精度逼近能力,可有效表征风电场一次调频复杂非线性响应特性。At present,frequency regulation of wind-farm-level grid connection faces significant challenges.In order to reduce modeling complexity,using actual operation data of grid connection points of the wind farm,a time-domain differential dynamic modeling method for primary frequency regulation characteristics under all operation conditions within the output range of wind farms is proposed.Firstly,the mechanism is analyzed and the input and output variables are determined to complete data sampling and processing.Then,discrete condition sampling and nonlinear evaluation methods are proposed,and a subspace model under single operation condition is established to evaluate the dynamic nonlinearity of frequency regulation response under all operation conditions through gap measurement.Thereafter,in order to efficiently approximate the complex nonlinear frequency regulation characteristics under all operation conditions,a hybrid finite difference regression vector is constructed,and its stretched differential dynamic space is convexly divided.Modeling samples are evenly selected in each sub operation domain,and long and short-term memory neural network with attention mechanism is used to model under all operation conditions.Finally,through the verification of the frequency regulation operation data of grid connection points of the wind farm,it shows that the proposed finite difference dynamic modeling method under all operation conditions has the approximation ability of arbitrary precision,and can effectively characterize the complex nonlinear response characteristics of primary frequency regulation for the wind farm.

关 键 词：风电场 一次调频 有限差分域划分 神经网络 子空间辨识 间隙测度 

分 类 号：TM614[电气工程—电力系统及自动化]