SPNCC与一维双通道CNN-LSTM相结合的变压器局部放电故障音频检测

PDF下载HTML

Title:

Audio detection of transformer partial discharge fault based on SPNCC and one-dimensional dual-channel CNN-LSTM

作者:

翟学明¹, 郭嘉¹, 翟羽佳²

1. 华北电力大学 控制与计算机工程学院, 河北 保定 071003;
2. 国网石家庄供电公司, 河北 石家庄 050000

Author(s):

ZHAI Xueming¹, GUO Jia¹, ZHAI Yujia²

1. School of control and computer engineering, North China Electric Power University, Baoding 071003, China;
2. State Grid Shijiazhuang Electric Power Company, Shijiazhuang 050000, China

关键词:

变压器局部放电; 变压器故障检测; 智能电网; 小波包分解; 简化版幂律归一化倒谱系数; Gammatone滤波器; 卷积神经网络; 长短时记忆网络

Keywords:

transformer partial discharge; transformer fault detection; smart grid; wavelet packet decomposition; simple power normalized cepstral coefficient; Gammatone filter; convolutional neural network; long short-term memory

分类号:

TP183；TM41

DOI:

10.11992/tis.202110036

摘要:

变压器运行时产生的声音信号包含了丰富的状态信息，可作为变压器故障诊断的重要依据。为提高变压器故障音频诊断效果，首先将变压器运行时现场采集的声音信号分为工作环境噪声信号、正常工作音频信号以及局部放电故障音频信号；然后通过卷积神经网络分类、小波包分解以及巴特沃斯带通滤波的方法去除原始音频信号中的非稳态环境噪声和短时稳态环境噪声信号；并建立了基于简化版幂律归一化倒谱系数特征的一维双通道卷积神经网络?长短时记忆网络的变压器局部放电故障识别模型。通过现场采集某500 kV变电站不同自然环境下的变压器运行声音信号与故障仿真实验，验证了提出的局部放电故障识别模型的可行性，相较于传统的音频故障诊断方法，提出的模型具有更快的收敛速度，更好的故障识别准确率与故障分类准确率。

Abstract:

The sound signals generated during transformer operation contain abundant state information, which can be used as an important basis for transformer fault diagnosis. In order to improve the audio diagnosis effect of transformer fault, firstly, the sound signals collected in the field during transformer operation are divided into working environment noise signals, normal working audio signals and partial discharge fault audio signals. Then, the non-stationary environmental noise and short-term stationary environmental noise signals in the original audio signals are removed by the methods of convolutional neural network classification, wavelet packet decomposition and Butterworth band-pass filtering. And a one-dimensional two-channel convolutional neural network-long short-term memory (CNN-LSTM) model for transformer partial discharge fault identification is established based on the simple power normalized cepstral coefficient (SPNCC) feature. By collecting sound signals from a 500 kV substation transformer operation under different natural environments and the fault simulation experiments, the feasibility of the proposed partial discharge fault recognition model is verified. Compared with traditional audio fault diagnosis methods, the proposed model has a faster convergence speed, better fault identification accuracy and fault classification accuracy.