Series arc fault detection based on continuous wavelet transform and DRSN-CW with limited source data.

Area of Science:

• Electrical Engineering
• Artificial Intelligence
• Power Systems

Background:

• Series arc faults in indoor power distribution systems pose significant fire risks due to extreme temperatures.
• Deep learning models are effective for fault diagnosis but are hindered by limited public datasets for arc fault detection.

Purpose of the Study:

• To develop an effective arc fault detection method addressing the challenge of scarce data.
• To improve the accuracy and reliability of arc fault detection in indoor power distribution systems.

Main Methods:

• Utilized continuous wavelet transform (CWT) to extract grayscale image features from source data.
• Implemented data augmentation techniques to expand the dataset.
• Developed and applied a deep residual shrinkage network with channel-wise thresholding (DRSN-CW) for arc fault detection.

Main Results:

• The proposed DRSN-CW model achieved a highest arc fault detection accuracy of 98.92%.
• The average detection accuracy across tests was 97.72%.
• The method demonstrated excellent performance in identifying arc faults.

Conclusions:

• The novel method effectively detects arc faults using CWT and DRSN-CW, overcoming data limitations.
• This approach offers a viable and high-performing solution for arc fault detection.
• Provides a new strategy for enhancing electrical fire safety in power distribution systems.