Joint processing technology of laser radar and optical image for power distribution

Area of Science:

• Electrical Engineering
• Computer Science
• Artificial Intelligence

Background:

• Traditional power distribution systems struggle with fault identification due to unpredictable loads and complex fault propagation, leading to false alarms and slow response times.
• Existing methods lack the precision required for efficient maintenance planning in modern, complex grid environments.

Purpose of the Study:

• To develop an advanced fault identification and localization system for power distribution networks.
• To overcome the limitations of traditional systems by enhancing accuracy, reducing response times, and improving precision.

Main Methods:

• The Multimodal Deep Feature Hybrid Deep Learning Model (MDF-HDL) integrates LiDAR, optical images, and sensor data for comprehensive fault identification.
• Deep learning layers create multimodal feature representations, enhanced by Kalman filtering for feature fusion.
• Decision trees, optimized with the Adam algorithm, refine classification results, while GIS mapping aids precise fault localization.

Main Results:

• The MDF-HDL model achieved high performance metrics: 98.91% accuracy, 98.7% precision, 98.3% recall, and 98.5% F1-score.
• The system demonstrated a rapid inference time of 12.5 milliseconds, indicating high efficiency.
• GIS mapping integration facilitated precise fault identification for effective maintenance planning.

Conclusions:

• The MDF-HDL model offers a dependable and effective solution for fault management in complex power grid contexts.
• By leveraging multimodal data and advanced algorithms, the system surpasses traditional constraints in fault identification and localization.
• The model's low computational complexity and high accuracy make it a valuable tool for modernizing power distribution infrastructure.