考虑惯量中心频率偏移的自编码器暂态稳定评估  

Transient Stability Assessment of Auto Encoder Considering Frequency Shift of Inertia Center

作　　者：赵冬梅[1] 王闯 谢家康 马泰屹 ZHAO Dongmei;WANG Chuang;XIE Jiakang;MA Taiyi(School of Electrical and Electronic Engineering,North China Electric Power University,Changping District,Beijing 102206,China)

机构地区：[1]华北电力大学电气与电子工程学院,北京市昌平区102206

出　　处：《电网技术》2022年第2期662-670,共9页Power System Technology

摘　　要：针对传统深度学习方法评估电力系统暂态稳定时没有考虑电力系统物理特性的问题,提出一种考虑系统惯量中心频率偏移量的电力系统暂态稳定评估方法。通过计算电力系统故障后的惯量中心频率偏移量,将样本进行分类,分别用堆叠稀疏自编码器进行训练。当系统网架结构发生改变时,采用迁移成分分析法结合惯量中心频率偏移量对分类器进行更新。通过新英格兰10机39节点系统上的仿真结果表明所提方法比传统深度学习方法及迁移学习方法精度更高、泛化性能更强。当部分同步向量测量单元缺失以及数据中含有噪声时也能取得很好的效果。Aiming at the problem that the traditional deep learning method does not consider the physical characteristics of power system when evaluating power system transient stability,a power system transient stability evaluation method considering the frequency offset of the system inertia center is proposed.By calculating the frequency offset of the inertia center after the power system faults,the samples are classified and trained with the stacked sparse auto encoder.When the grid structure of the system changes,the method of transfer component analysis combined with the inertia center frequency offset is used to update the classifier.Simulation results on the New England 10 machine 39 bus system show that the proposed method has higher accuracy and stronger generalization performance than the traditional deep learning method and the transfer learning method.The proposed method still achieves good results when some synchronous vector measurement units are missing or there is noise in the data.

关 键 词：深度学习 电力系统 惯量中心频率 暂态稳定 堆叠稀疏自编码器 

分 类 号：TM721[电气工程—电力系统及自动化]