A novel arc detection and identification method in pantograph-catenary system based on deep learning.

Area of Science:

• Electrical Engineering
• Computer Science
• Artificial Intelligence

Background:

• Arc detection is vital for power system safety, preventing fires and equipment damage.
• Traditional methods lack accuracy and efficiency, especially in real-time monitoring.
• Complex operational environments pose challenges for existing arc detection techniques.

Purpose of the Study:

• To propose an advanced deep learning method for accurate and efficient arc detection.
• To enhance the Inception V3 model for improved performance in power system monitoring.
• To address limitations of traditional arc detection methods in terms of accuracy and complexity.

Main Methods:

• Developed Arc Multi-Scene Detection (ArcMSD) using a modified Inception V3 deep learning model.
• Incorporated guided anchor, attention, and upsampling mechanisms into the Inception module.
• Utilized preprocessed images from Integrated Supervision and Control System (ISCS) videos for training.

Main Results:

• The ArcMSD model achieved a mean Average Precision (mAP) of 95.4%.
• The enhanced Inception V3 demonstrated superior arc feature discrimination and reduced overfitting.
• The guided anchor mechanism improved detection efficiency for arcs of varying sizes.

Conclusions:

• The proposed ArcMSD method significantly outperforms existing models in arc detection accuracy.
• Deep learning, particularly the enhanced Inception V3, offers a robust solution for power system safety.
• The method is effective for real-time monitoring and efficient operation of power systems.