摘要
针对供配电网络中变压器设备数量众多、故障损失巨大、不能及时有效地实现故障检测和预报等行业现状,利用大数据方法研究分析了众多变压器的实时运行数据,提出基于深度学习的变压器故障检测方法,详细介绍了变压器监测数据预处理方法及步骤;首先变压器实时运行数据经过分类、组合等预处理运算,转换成多维空间的状态数据,最后进一步将多维空间状态数据拟合成多段状态变迁的曲线,作为深度学习网络的输入训练样本;基于简洁高效的经典开源的AlexNet卷积神经网络模型,搭建了基于tensorflow架构的深度学习训练平台,实现了基于深度学习网络的变压器在线故障检测,系统运行效果表明该故障检测方法的有效性和实用性。
In view of the large number of transformer equipment,the huge loss of fault,and the failure detection and prediction can not be realized in time and effectively in the power supply and distribution network,this paper studies and analyzes the real-time operation data of many transformers by using the bulk data method,puts forward the transformer failure detection method based on deep learning,and introduces the pretreatment method and steps of transformer monitoring data in detail.At first,the real-time operation data of transformer is transformed into multi-dimensional state data by pre-processing operations such as classification and combination.Finally,the multi-dimensional state data is further fitted into multi-stage state transition curve as the input training sample of deep learning network.Based on the simple and efficient classic open-source AlexNet convolutional neural network model,a deep learning training platform based on Tensorflow architecture is built,and the transformer online fault detection based on deep learning network is realized.The system operation results show that the fault detection method is effective and practical..
作者
童国锋
朱梅
Tong Guofeng;Zhu Mei(Keqiao district power supply branch,Shaoxing Electric Power Bureau,Shaoxing 312030,China;Information College,Zhejiang University of Science and Technology,Hangzhou 310023,China)
出处
《计算机测量与控制》
2020年第9期65-68,共4页
Computer Measurement &Control
基金
国家自然科学基金(61075062)。
关键词
深度学习
故障检测
密度图像
曲线拟合
deep learning
fault detection
density image
curve fitting