Federated Learning-Based Insulator Fault Detection for Data Privacy Preserving

Area of Science:

• Electrical Engineering
• Computer Science
• Artificial Intelligence

Background:

• Insulators are critical components in power distribution networks, and their faults can disrupt operations.
• Manual fault detection is inefficient and prone to errors.
• Vision sensors offer automated detection, but centralized processing raises privacy concerns.

Purpose of the Study:

• To propose a privacy-preserving insulator fault detection method using federated learning.
• To address the limitations of centralized data processing in vision-based insulator fault detection.
• To maintain high detection accuracy while protecting data privacy.

Main Methods:

• Developed a federated learning framework for insulator fault detection.
• Constructed a dedicated insulator fault detection dataset.
• Trained Convolutional Neural Network (CNN) and Multi-Layer Perceptron (MLP) models within the federated learning environment.

Main Results:

• The proposed federated learning method achieves insulator anomaly detection accuracy exceeding 90%.
• The method effectively protects data privacy during model training.
• Experimental results validate the framework's applicability and privacy-preserving capabilities.

Conclusions:

• Federated learning provides an effective solution for privacy-preserving insulator fault detection.
• The proposed method balances high detection accuracy with robust data security.
• This approach enhances the reliability and safety of power distribution networks.