Anti-Interference Deep Visual Identification Method for Fault Localization of Transformer Using a Winding Model.

Area of Science:

• Electrical Engineering
• Artificial Intelligence
• Power Systems

Background:

• Ubiquitous Power Internet of Things (UPIoTs) drive demand for advanced intelligent monitoring and diagnostics.
• Traditional fault diagnostic methods for power equipment face challenges in interference-heavy environments.

Purpose of the Study:

• To propose a robust fault identification method for power equipment operating in interference environments.
• To enhance diagnostic accuracy, anti-interference ability, and fault tolerance in power equipment monitoring.

Main Methods:

• Developed a diagnostic method combining deep Convolutional Neural Network (CNN), specifically MobileNet-V2, with Digital Image Processing (DIP).
• Utilized data visualization theory on transformer frequency response curves to create a dataset, followed by image augmentation.
• Established a spatial-probabilistic mapping relationship based on traditional Frequency Response Analysis (FRA) for comparative analysis.

Main Results:

• The proposed CNN-DIP method demonstrated higher diagnostic accuracy compared to traditional methods.
• The method exhibited superior anti-interference capabilities as interference magnitude increased.
• Experimental analysis confirmed the enhanced fault tolerance of the CNN-DIP approach.

Conclusions:

• The deep visual identification (CNN-DIP) method offers a significant advancement in power equipment fault diagnosis.
• This approach provides a more reliable and accurate solution for intelligent monitoring in challenging industrial settings.