Research on Low-Voltage Arc Fault Based on CNN-Transformer Parallel Neural Network with Threshold-Moving Optimization

Area of Science:

• Electrical Engineering
• Artificial Intelligence
• Safety Systems

Background:

• Low-voltage arc faults pose significant fire and shock risks.
• Effective arc fault detection is crucial for preventing accidents and protecting property.
• Existing methods may struggle with diverse load characteristics during arc faults.

Purpose of the Study:

• To develop an efficient and accurate method for low-voltage arc fault recognition.
• To leverage deep learning for enhanced arc fault detection.
• To improve the safety of electrical systems in residential and industrial settings.

Main Methods:

• A parallel neural network architecture combining Convolutional Neural Networks (CNNs) and Transformers.
• CNNs process low-frequency current components; Transformers process high-frequency components.
• Softmax classification is used for supervised learning on concatenated features.

Main Results:

• Achieved a high detection accuracy of 99.74%.
• Demonstrated a reduced erroneous judgment rate using a threshold-moving method.
• The proposed network shows improved recognition efficiency due to its lean structure.

Conclusions:

• The parallel neural network effectively detects low-voltage arc faults.
• The hybrid approach combines CNN and Transformer strengths for robust recognition.
• This method offers a promising solution for enhancing electrical safety systems.