基于分布式光纤传感系统的有载变压器故障检测技术研究  

On-load transformer fault detection based on distributed optical fiber sensing system

作　　者：董贯雷 姜晓东 孙鹏 杨光 耿俊琪 王家文 渠帅 黄胜 王晨[2] 尚盈 DONG Guanlei;JIANG Xiaodong;SUN Peng;YANG Guang;GENG Junqi;WANG Jiawen;QU Shuai;HUANG Sheng;WANG Chen;SHANG Ying(Zibo Power Supply Company,State Grid Shandong Electric Power Company,Zibo 255000,China;Laser Institute,Qilu University of Technology(Shandong Academy of sciences),Jinan 250014,China)

机构地区：[1]国网山东省电力公司淄博供电公司,山东淄博255000 [2]齐鲁工业大学(山东省科学院)激光研究所,山东济南250104

出　　处：《山东科学》2023年第5期52-59,共8页Shandong Science

基　　金：国网山东省电力公司科技项目(520603220003)。

摘　　要：提出基于分布式光纤传感技术的人工神经网络有载变压器故障检测预报模型,通过人工模拟变压器的故障状态及正常运行状态,并通过k-means SMOTE数据扩充方法,可以有限扩充少量故障数据集,使故障数据量可以和正常数据量达到一致,将扩充后的故障数据与正常运行的数据一起送入长短期记忆卷积神经网络(convolutional neural networks long short term memory, CNN-LSTM)识别模型,最终可以将故障的识别率提升到100%,这对采用分布式光纤传感技术在有载变压器故障识别系统上的发展具有重要意义。This paper proposes an artificial neural network-based fault detection and prediction model for on-load transformers using distributed fiber optic sensing technology.By artificially simulating the fault and normal operating states of transformers and using the k-means synthetic minority oversampling technique data expansion method,a small number of fault datasets can be limitedly expanded so that the amount of fault data can be consistent with that of normal data.Therefore,the expanded fault data and normal operation data can be input into the convolutional neural networks long short term memory identification model.Finally,the fault recognition rate can be increased to 100%,which has significant implications for the development of fault recognition systems for on-load transformers based on distributed fiber optic sensing technology.

关 键 词：神经网络 分布式声传感器 有载变压器 故障检测 SMOTE算法 模式识别 数据增强 

分 类 号：TN247[电子电信—物理电子学]