摘要
针对常规迁移学习辨识网络需要大量历史辅助数据而影响参数辨识准确性的问题,提出考虑小样本下两阶段迁移学习的调相机暂态电抗参数在线辨识方法。首先,获得不同运行工况下的暂态电抗参数变化规律,利用支持向量机(SVM)和反向传播(BP)神经网络算法获得辅助数据。其次,基于电抗参数变化规律、故障下实时变化的暂态电抗目标数据及辅助数据,采用深度稀疏自编码器(DSAE)与反向传播神经网络进行两阶段特征迁移学习。最后,对不同电网故障下的暂态电抗进行在线辨识,并进行比较验证。结果表明,基于两阶段迁移学习的DSAE与BP神经网络在辨识调相机暂态电抗参数时具有更高的辨识精度。
Aiming at the problem that a conventional transfer learning identification network requires a large amount of historical auxiliary data,which affects the accuracy of parameter identification,an online identification method for transient reactance parameters of condensers considering the two-stage transfer learning with limited samples is proposed.Firstly,the variation laws of transient reactance parameters under different operation conditions are obtained,and the auxiliary data is acquired by using the algorithm such as support vector machine(SVM)and back propagation(BP)neural network.Secondly,based on the variation laws of reactance parameters,the target data and auxiliary data of real-time variable transient reactance under faults,the two-stage feature transfer learning is conducted by adopting the deep sparse automatic encoder(DSAE)and BP neural network.Finally,the online identification of transient reactance under different power grid faults and comparative verification are conducted.The results show that the DSAE and BP neural network based on the two-stage transfer learning have higher identification accuracy in identifying the transient reactance parameters of condensers.
作者
袁彬
李辉
向学位
曾韵竹
谭宏涛
傅楚频
YUAN Bin;LI Hui;XIANG Xuewei;ZENG Yunzhu;TAN Hongtao;FU Chupin(State Key Laboratory of Power Transmission Equipment&System Security and New Technology(Chongqing University),Chongqing 400044,China)
出处
《电力系统自动化》
EI
CSCD
北大核心
2023年第14期149-156,共8页
Automation of Electric Power Systems
基金
国家高技术船舶科研项目(MC-202025-S02)。
关键词
调相机
暂态电抗
参数辨识
故障
迁移学习
condenser
transient reactance
parameter identification
fault
transfer learning
