基于FNN的变压器绕组振动模型建立及研究  

Establishment and Research of Transformer Winding Vibration Model Based on Fourier Neural Network

作　　者：刘石 韩丹 杨毅 郑婧[3] 黄海[3] 李灵至 LIU Shi;HAN Dan;YANG Yi;ZHENG Jing;HUANG Hai;LI Lingzhi(Electric Power Research Institute of Guangdong Power Grid Co.,Ltd.,Guangzhou,Guangdong 510080,China;Guangdong Diankeyuan Energy Technology Co.,Ltd.,Guangzhou,Guangdong 510080,China;Department of Instrumentation Science and Engineering,Zhejiang University,Hangzhou,Zhejiang 310027,China)

机构地区：[1]广东电网有限责任公司电力科学研究院,广东广州510080 [2]广东电科院能源技术有限责任公司,广东广州510080 [3]浙江大学仪器科学与工程学系,浙江杭州310027

出　　处：《广东电力》2018年第8期62-68,共7页Guangdong Electric Power

基　　金：广东电网有限责任公司科技项目(GDKJQQ20153006)

摘　　要：基于傅里叶神经网络(Fourier neural network,FNN)结构的单输入、输出Hammerstein模型,提出了一种适用于变压器绕组振动系统的非线性建模方法。FNN和自回归滑动平均模型(auto-regressive and moving average model,ARMA)(分别作为模型中的非线性静态模块和线性动态模块)采用前向更新策略及最速下降法对模型进行训练,确定模型参数。该建模方法在实际110 kV变压器的绕组振动系统的建模及绕组振动波形预测中显示了较高的有效性及准确性。此外,还研究和分析了绕组故障对模型特性的影响机制,提出基于模型延迟阶数的绕组故障特征及其提取方法,并将该方法及其特征量应用于实际110 kV变压器的绕组故障实验中,所得结果表明该特征量可有效反映变压器绕组的机械结构变化。Based on the single input and output Hammerstein model of Fourier neural network （FNN） structure, this paper presents a kind of non-linear modeling method suitable for transformer winding vibration system. Respectively as the non-linear static module and the linear dynamic module of the model. The FNN model and the auto-regressive and moving average （ARMA） model both adopt the forward direction update strategy and the steepest descent method for training the model and determining parameters of the model. This method has better effectiveness and accuracy in actual modeling for the winding vibration system of 110 kV transformer and forecasting on winding vibration waveforms. In addition, the paper studies and analyzes influencing mechanism of winding fault on model characteristics, and proposes a method for winding fault characteristic extraction based on model delay orders. Application of this fault characteristic extraction method and its characteristic quantities in an actual winding fault test for 110 kV transformer has proved fault characteristics can effectively reflect changes of mechanical structure of transformer winding.

关 键 词：绕组 振动 非线性模型 傅里叶神经网络 故障特征提取 HAMMERSTEIN模型 

分 类 号：TH113.1[机械工程—机械设计及理论] TM411[电气工程—电器]