基于时频域融合和ECA-1DCNN的航空串联故障电弧检测

Aviation Series Arc Fault Detection Method Based on Time Frequency Domain Fusion and ECA-1DCNN

为了快速准确地检测航空交流线路中出现的串联故障电弧,提出了一种基于时频域融合和加入高效注意力机制(efficient channel attention, ECA)的一维卷积神经网络(one-dimensional convolutional neural network, 1DCNN)的故障检测算法。首先,搭建航空交流电弧故障实验平台,负载选择多类型、多参数值进行电流信号的采集;其次,为了保留更多的故障信息,分析其特征频段,经过大量数据验证,航空串联电弧在发生时,1 000~4 000 Hz分量具有一定的占比,因此将原始信号与特征频段进行融合,融合后的一维数据作为模型输入;最后,搭建ECA-1DCNN检测模型,进行训练,并通过K折交叉验证模型的有效性,得到测试集平均准确率为97.96%。该方法网络层数较少,计算快速,避免了复杂时频域计算过程,较为智能,对航空串联电弧检测装置的研究提供了理论参考。

A fault detection algorithm for detecting serial arc faults in aviation communication lines has been proposed based on time-frequency domain fusion and the inclusion of efficient channel attention(ECA)in a one-dimensional convolutional neural network(1DCNN).Firstly,an aviation AC arc fault experimental platform was established,and loads of various types and parameter values were selected for the collection of current signals.Secondly,in order to retain more fault information,the time-domain signal was subject to a fast Fourier transform to observe the frequency spectrum,analyze its characteristic frequency bands,and after a large amount of data validation,it was found that when aviation serial arcing occurs,the 1000~4000 Hz components have a certain proportion.Therefore,the original signal was fused with the characteristic frequency band,and the fused one-dimensional data was used as the input to the model.Finally,an ECA-1DCNN detection model was constructed and trained,and the validity of the model was verified through K-fold cross-validation to obtain an average accuracy of 97.96%on the test set.The method has a low number of network layers,quick computation,avoids complex time-frequency domain calculation processes,and is more intelligent,providing theoretical reference for the development of aviation serial arc detection devices.