<p>Partial discharge (PD) monitoring is critical in ensuring the reliability and longevity of electrical insulation systems in high-voltage equipment. However, the accurate detection of PD signals is often hindered by strong interference from noise sources in industrial environments. This research article introduces a new deep learning-based denoising framework that integrates a convolutional neural network (CNN)- based denoising autoencoder (CDeAE) for effectively denoising acoustic PD signals. The proposed method leverages the feature extraction capabilities of CNNs and the noise reduction efficiency of autoencoders to enhance signal clarity while preserving essential PD characteristics. The framework is evaluated using measured, and simulated PD signals contaminated with Gaussian white noise at varying intensity levels. Performance metrics such as signal-to-noise ratio (<InlineEquation ID="IEq1"> <InlineMediaObject> <ImageObject Color="BlackWhite" FileRef="11760_2025_4597_Article_IEq1.gif" Format="GIF" Height="16" Rendition="HTML" Resolution="72" Type="Linedraw" Width="44" /> </InlineMediaObject> <EquationSource Format="TEX">${\text{SNR}}_{\text{D}}$</EquationSource> <EquationSource Format="MATHML"><math> <msub> <mtext>SNR</mtext> <mtext>D</mtext> </msub> </math></EquationSource> </InlineEquation>), peak signal-to-noise ratio (PSNR), root mean square error (RMSE), and reduction to noise level (<InlineEquation ID="IEq2"> <InlineMediaObject> <ImageObject Color="BlackWhite" FileRef="11760_2025_4597_Article_IEq2.gif" Format="GIF" Height="16" Rendition="HTML" Resolution="72" Type="Linedraw" Width="23" /> </InlineMediaObject> <EquationSource Format="TEX">${\text{R}}_{\text{nl}}$</EquationSource> <EquationSource Format="MATHML"><math> <msub> <mtext>R</mtext> <mtext>nl</mtext> </msub> </math></EquationSource> </InlineEquation>) are employed to assess its effectiveness. Comparative analysis demonstrates that the proposed CDeAE framework outperforms the state-of-the-art denoising techniques. Particularly, it outperforms existing methods by achieving an SNR improvement of 23.735 dB, a reduced RMSE of 0.1024, a PSNR of 20.79 dB and a <InlineEquation ID="IEq3"> <InlineMediaObject> <ImageObject Color="BlackWhite" FileRef="11760_2025_4597_Article_IEq2.gif" Format="GIF" Height="16" Rendition="HTML" Resolution="72" Type="Linedraw" Width="23" /> </InlineMediaObject> <EquationSource Format="TEX">${\text{R}}_{\text{nl}}$</EquationSource> <EquationSource Format="MATHML"><math> <msub> <mtext>R</mtext> <mtext>nl</mtext> </msub> </math></EquationSource> </InlineEquation> of 90.0021 at 15 dB input SNR for the different sensors. The proposed method significantly enhances the accuracy and reliability of PD detection, making it a promising solution for real-time monitoring of insulation health in power systems.</p>

A hybrid deep learning framework for denoising acoustic partial discharge signals

期刊论文
发表日期：

2025年9月5日

查看全文

Chandan Kumar,
Biswarup Ganguly,
Debangshu Dey,
Saibal Chatterjee

摘要

Partial discharge (PD) monitoring is critical in ensuring the reliability and longevity of electrical insulation systems in high-voltage equipment. However, the accurate detection of PD signals is often hindered by strong interference from noise sources in industrial environments. This research article introduces a new deep learning-based denoising framework that integrates a convolutional neural network (CNN)- based denoising autoencoder (CDeAE) for effectively denoising acoustic PD signals. The proposed method leverages the feature extraction capabilities of CNNs and the noise reduction efficiency of autoencoders to enhance signal clarity while preserving essential PD characteristics. The framework is evaluated using measured, and simulated PD signals contaminated with Gaussian white noise at varying intensity levels. Performance metrics such as signal-to-noise ratio ( ${\text{SNR}}_{\text{D}}$ ${SNR}_{D}$ ), peak signal-to-noise ratio (PSNR), root mean square error (RMSE), and reduction to noise level ( ${\text{R}}_{\text{nl}}$ $R_{nl}$ ) are employed to assess its effectiveness. Comparative analysis demonstrates that the proposed CDeAE framework outperforms the state-of-the-art denoising techniques. Particularly, it outperforms existing methods by achieving an SNR improvement of 23.735 dB, a reduced RMSE of 0.1024, a PSNR of 20.79 dB and a ${\text{R}}_{\text{nl}}$ $R_{nl}$ of 90.0021 at 15 dB input SNR for the different sensors. The proposed method significantly enhances the accuracy and reliability of PD detection, making it a promising solution for real-time monitoring of insulation health in power systems.