基于深度学习的变换器宽范围暂态等效建模  

Dynamic Equivalent Modeling for Power Converter in Wide-range by Using Deep Learning Method

作　　者：胡博 杨超 田小蕾 李云路 杨俊友[2] 崔嘉[2] HU Bo;YANG Chao;TIAN Xiaolei;LI Yunlu;YANG Junyou;CUI Jia(State Grid Liaoning Electric Power Supply Co.,Ltd.,Shenyang 110004,Liaoning,China;School of Electrical Engineering,Shenyang University of Technology,Shenyang 110870,Liaoning,China)

机构地区：[1]国网辽宁省电力有限公司,辽宁沈阳110004 [2]沈阳工业大学电气工程学院,辽宁沈阳110870

出　　处：《电气传动》2022年第9期26-31,共6页Electric Drive

基　　金：辽宁省“兴辽英才计划”(XLYC1902090)。

摘　　要：当电力变换器的内部参数及拓扑未知时,现有的基于频率扫描阻抗建模方法仅能保证所建立的暂态模型在单一工作点有效。为了使所建立的暂态模型在宽范围内有效,提出了一种基于深度学习方法的变换器暂态等效建模方法。首先,对变换器黑盒建模问题与深度循环神经网络的等效性进行研究。然后,提出基于门控单元循环神经网络的变换器黑盒宽范围等效建模方法。最后,在变换器接入点大扰动下、多工作点下进行仿真实验,验证了所提出的方法在宽范围内的暂态过程均具有等效性。When the internal parameters and topology of grid-connected power converter is unknown in some practical applications,the existing frequency scanning based impedance modeling method can only maintain the dynamic equivalence at one operating point.To ensure the effectiveness of dynamic model in a wide operating range,a deep learning based dynamic equivalent modeling method was proposed.Firstly,the equivalence between black-box modeling problem of power converter and deep recurrent neural network was studied.Secondly,a black box equivalent modeling method based on gate recurrent unit recurrent neural network was proposed to solve the wide range modeling issue.At last,under large perturbation and multi-operating points,simulations were implemented to validate the equivalence in a wide range of dynamic process of the proposed method.

关 键 词：电力变换器 暂态建模 深度学习 神经网络 等效建模 

分 类 号：TM933[电气工程—电力电子与电力传动]