基于STFT和CNN-Attention的配电终端采集模块故障诊断研究

Fault Diagnosis Study of Power Distribution Terminal Acquisition ModuleBased on STFT and CNN-Attention

针对复杂工况运行环境下配电终端采集模块故障类型难以识别的问题,提出一种基于短时傅里叶变换(STFT)、卷积神经网络和注意力机制(CNN-Attention)的配电终端采集模块故障诊断方法。首先,分析配电终端采集模块不同故障类型会产生的对应故障数据,建立故障数据集。然后,基于STFT提取故障数据的故障时频特征以形成时频图,采用CNN-Attention模型对时频图进行故障诊断与匹配。算例分析表明,CNN-Attention的故障检测准确率为97.31%,相较于CNN和极限学习机(ELM)模型,故障诊断准确率分别提升了1.22%和4.4%。Attention机制能够有效解决CNN在特征提取时产生的冗余信息导致模型训练慢、难以收敛的问题。该研究实现了配电终端采集模块具体故障类型的准确识别,能为后续配电终端的运维提供参考。

Aiming at the problem that it is difficult to recognize the fault types of power distribution terminal acquisition module under the operating environment of complex working conditions,a fault diagnosis method of power distribution terminal acquisition module based on short-time Fourier transform(STFT),convolutional neural network and attention mechanism(CNN-Attention) is proposed.Firstly,the corresponding fault data generated by different fault types of the power distribution terminal acquisition module are analyzed to establish a fault data set.Then,the fault time-frequency features of the fault data are extracted based on STFT to form a time-frequency diagram,and the CNN-Attention model is used to diagnose and match the faults on the time-frequency diagram.Case analysis shows that CNN-Attention fault detection accuracy is 97.31%,compared with CNN and extreme learning machine(ELM) models,the fault diagnosis accuracy is improved by 1.22% and 4.4%,respectively.Attention mechanism can effectively solve the problem of slow model training and difficulty in convergence due to the redundancy of information generated by the CNN during feature extraction.This study realizes the accurate identification of specific fault types in the collection module of power distribution terminals,which can provide reference for the subsequent operation and maintenance of power distribution terminals.