融合物理与数据知识的电力系统扰动后频率在线快速计算方法  

On-line Fast Frequency Calculation After Power System Disturbance Based on Fusion of Physics and Data Knowledge

作　　者：张涵 王程 毕天姝 ZHANG Han;WANG Cheng;BI Tianshu(State Key Laboratory of Alternate Electrical Power System With Renewable Energy Sources(North China Electric Power University),Changping District,Beijing 102206,China)

机构地区：[1]新能源电力系统国家重点实验室(华北电力大学),北京市昌平区102206

出　　处：《电网技术》2022年第11期4325-4335,共11页Power System Technology

摘　　要：针对电力系统扰动后频率响应计算问题,该文基于门控循环单元神经网络提出一种融合物理与数据知识的频率在线计算方法,以实现频率快速精准计算。该方法以同步电源惯性时间常数等影响频率响应的主导因素作为经典“黑箱”机器学习方法的基本输入特征量,并进一步在“黑箱”方法中嵌入频率响应相关物理知识,通过基本输入特征量和所嵌入物理知识形成新的输入特征量并用于模型训练。该方法能够提高小样本场景下的模型泛化能力和抗噪能力,并且增强其可解释性。采用新英格兰10机39节点系统作为仿真算例,通过与电力系统仿真器(power system simulator for engineering,PSS/E)中的仿真结果相对比,证明所提方法能够快速、准确地计算电力系统扰动后频率响应曲线。Aiming at the problem of frequency response calculation after the power system disturbance, this paper proposes a new method of online frequency calculation based on the gated recurrent unit neural network that combines physics and data knowledge to achieve fast and accurate frequency calculation. This method takes the dominant factors that affect the frequency response, such as the inertia time constant of the synchronous power supply, as the basic input feature quantity of the classic "black box" machine learning method. Further it embeds the physical knowledge of the frequency response in the "black box" method. Through the basic input feature quantity and the embedded physical knowledge, it forms a new input feature quantity and uses it for model training. This method improves the generalization ability and anti-noise ability of the model in a small sample scene, and enhances its interpretability. Using a New England 10-machine 39-bus system as a simulation example, compared with the simulation results in PSS/E, it is proved that the proposed method can quickly and accurately calculate the frequency response curve after the power system disturbance.

关 键 词：频率稳定 频率在线计算 物理–数据联合驱动 门控循环单元神经网络 

分 类 号：TM721[电气工程—电力系统及自动化]