Series Arc Fault Detection Based on Multimodal Feature Fusion

Area of Science:

• Electrical Engineering
• Power Systems
• Signal Processing

Background:

• Complex load types in low-voltage distribution systems challenge traditional series arc fault detection.
• Existing methods struggle with accurately identifying series arc faults due to signal complexity.

Purpose of the Study:

• To propose an advanced arc fault detection method utilizing multimodal feature fusion.
• To enhance the accuracy and performance of series arc fault identification in low-voltage distribution systems.

Main Methods:

• Extraction of diverse signal features: time-domain, frequency-domain, wavelet packet energy, and time-spectrum images.
• Application of machine learning algorithms for feature preprocessing, prioritization, and quality improvement.
• Development of a hybrid deep learning model combining 1D Convolutional Networks and Deep Residual Shrinkage Networks for detection.

Main Results:

• The proposed multimodal feature fusion method significantly improves arc fault detection accuracy.
• The integrated deep learning model demonstrates superior performance compared to single-mode feature-based methods.
• Feature prioritization enhances the quality of input data for the detection model.

Conclusions:

• Multimodal feature fusion is an effective strategy for improving series arc fault detection in complex low-voltage systems.
• The combination of advanced signal processing and deep learning offers a robust solution for electrical safety.
• This method provides a more reliable approach to identifying arc faults, enhancing system safety and stability.