基于S变换时频谱和KHA-CNN的换流变故障声纹识别
Voiceprint Recognition of Converter Transformer Faults Based on S Transform Time-Frequency Spectrum and KHA-CNN
柴斌 1韦鹏 2宁复茂 1姚琪 3李辉3
作者信息
- 1. 国网宁夏电力有限公司超高压公司,宁夏银川 750011
- 2. 国网宁夏电力有限公司,宁夏银川 750001
- 3. 西安理工大学电气工程学院,陕西西安 710054
- 折叠
摘要
为准确进行换流变压器故障诊断,保证直流输电可靠性,提出一种基于S变换时频谱和KHA-CNN模式的换流变故障声纹识别方法.利用自适应补充集合经验模态分解算法(complete ensemble empirical mode decomposition with adaptive noise,CEEMDAN),实现换流变本体声纹信号与噪声的分离;通过S变换得到时频谱图,实现声纹信号的特征提取;通过磷虾群优化算法(krill herd algorithm,KHA)对卷积神经网络进行超参数寻优,将S时频谱图作为特征输入到KHA-CNN,实现故障诊断.研究结果表明:该方法对于换流变故障具有很好的识别效果,能为换流变故障诊断提供有效参考.
Abstract
Aiming to accurately diagnose converter transformer faults and ensure the reliability of DC transmission,this paper proposes a method for the voiceprint recognition of converter transformer faults based on S Transform time-frequency spectrum and KHA-CNN patterns.Firstly,the Complete Ensemble Empirical Mode Decomposition with Adaptive Noise(CEEMDAN)algorithm is used to separate the voiceprint signal and noise of the converter body;secondly,the time-frequency spectrum is obtained by the S-transform to achieve extraction of features of the voiceprint signal.Finally,the Krill Herd Algorithm(KHA)is used to optimize the convolutional neural network with hyperparameters,and the S-time-frequency spectrogram is used as the feature input to the KHA-CNN for fault diagnosis.The research results show that the method has good recognition effect for converter faults and provides effective reference for converter fault diagnosis.
关键词
换流变压器/声纹信号/S变换时频谱图/磷虾群优化算法/卷积神经网络Key words
converter transformer/voiceprint signal/S transform time-frequency spectrum/krill herd algorithm/convolutional neural network引用本文复制引用
基金项目
国家自然科学基金(51779206)
宁夏自然科学基金(2022AAC03631)
出版年
2024
NETL
NSTL
万方数据