A Review of Physics-Informed and Data-Driven Surrogate Models for Power Transformer Fault Diagnosis

Area of Science:

• Electrical Engineering
• Artificial Intelligence
• Power Systems

Background:

• Power transformers are vital for grid reliability and security.
• Traditional fault diagnosis methods face challenges with complex power systems.
• Surrogate models show promise for intelligent transformer condition monitoring.

Purpose of the Study:

• To systematically review surrogate models for power transformer fault diagnosis.
• To establish a classification framework for surrogate modeling approaches.
• To analyze the performance, advantages, and limitations of various surrogate models.

Main Methods:

• Comprehensive literature review on surrogate models in transformer fault diagnosis.
• Classification of models based on type, strategy, data, and application.
• Comparative analysis of principles, conditions, and performance characteristics.
• Discussion of diagnostic accuracy, stability, generalization, interpretability, and efficiency.

Main Results:

• Surrogate models excel in high-dimensional nonlinear mapping and rapid prediction.
• Analysis reveals varying performance across different surrogate modeling approaches.
• Identified challenges include small-sample learning and data quality dependence.

Conclusions:

• Surrogate models present a powerful, data-driven approach for intelligent transformer fault diagnosis.
• Future research should address multi-source fusion, hybrid learning, and standardization.
• This review provides guidance for developing reliable and interpretable diagnostic methods.