摘要
为了有效、快速评估公路交通工程中低压配电变压器的运行状态,提出基于深度学习的电力变压器运行状态评估新方法。该方法以变压器运行时振动信号为研究对象,先对振动信号进行预处理,后依据深度学习的无监督学习对振动信号进行特征学习,再以此特征作为BP神经网络的输入实现电力变压器振动信号特征识别,并将采用该方法与传统的BP神经网络方法的计算结果进行比较。结果表明,基于深度学习的电力变压器运行状态识别方法识别率平均可达97.64%,平均时间为4.91 s;基于传统BP神经网络方法的电力变压器运行状态识别方法的识别率平均只有84.88%,平均时间为21.41 s。由此说明了新方法的有效性,且为低压配电变压器运行状态评估和故障诊断提供一种新思路。
In order to make an effective and rapid evaluation of low power transformer in highway traffic engineering,a new approach,based on deep learning,is proposed.Vibration signal is studied in this method during operation of transformer.vibration signal is firstly preprocessed,and unsupervised learning based on deep learning is carried out on the preprocessed signal to obtain the characteristics.Then this feature is put as the input of BP neural network,and the power transformer vibration signal is identified.Compared with that of the traditional BP neural network method,the experimental results by this method show that the recognition rate of the power transformer running state method based on the deep learning is 97.64%on average,the calculation time is 4.91 s on average.The recognition rate of the power transformer running state method based on the traditional BP neural network is 84.88%on average,the calculation time is 21.41 s on average.It explains that the new method is effective.It provides a new idea for the operation state evaluation and fault diagnosis of low power transformer winding support node.
作者
袁源
庞荣
代东林
李响
韩坤林
许培振
YUAN Yuan;PANG Rong;DAI Donglin;LI Xiang;HAN Kunlin;XU Peizhen(China Merchants Chongqing Road Engineering Inspection Center Co.,Ltd.,Chongqing 400067;School of Information Science and Technology,Southwest Jiaotong University,Sichuan Chengdu 610031)
出处
《公路交通技术》
2020年第6期114-119,共6页
Technology of Highway and Transport
基金
重庆市社会事业与民生保障科技创新专项重点研发项目(cstc2017shms-zdyfX0007)。
关键词
深度学习
变压器
状态评估
故障诊断
deep learning
power transformer
state assessment
fault diagnosis
