基于AlexNet深度学习网络的串联故障电弧检测方法

Series arc fault detection method based on AlexNet deep learning network

在家用交流供配电系统中,接触松动等原因可能会导致故障电弧的发生,威胁用电系统的安全。线路发生串联故障电弧时的电流基本与正常运行时的电流大小一致,具有很强的隐蔽性。对此,首次提出用深度学习检测电流信号的方法来检测串联故障电弧,该方法只需将电流信号输入深度学习网络,由网络自主挖掘隐含在电流信号数据背后的特征,实现对串联故障电弧的识别。搭建实验平台,并用开关模拟发生正常电弧,分别采集电阻性负载、电感性负载和阻感性负载正常运行和发生串联故障电弧时的电流数据共7 200组。构建AlexNet卷积神经网络并做相应改进,用采集到的数据训练网络并测试,结果显示辨识平均准确率在85%以上,表明该方法能够较好的实现对串联故障电弧的检测。

In domestic alternating current power supply and distribution system,arc fault is caused by loose connect and other reasons and endangers the power system.The current that series arc fault generates is basically the same as the normal operation,so it is concealed.For this reason,an arc fault detection method using deep learning to detect current data is firstly proposed.This method only needs to input the current signal data to the deep learning network,the feature hidden behind the current signal data autonomously extracted by the network and the series arc fault is identified.A platform is set up and a key is used to simulate the normal arc.Normal current signals and series arc fault current signals of resistive load,inductance load and resistive-inductive load totally 7 200 groups are collected using the platform AlexNet convolutional neural network is constructed and improved accordingly,and trained and tested using the collected data.Experiments show that the average accuracy of identification is above 85%,which indicates that the series arc fault can be better detected.