Intelligent Diagnosis Method for Transformer Insulation Aging Based on Multimodal Data Fusion

This study presents an intelligent diagnostic approach for transformer insulation aging using multimodal data fusion and Graph Neural Networks (GNN). By integrating oil chromatography, partial discharge signals, and infrared thermography data into a heterogeneous information graph, the method captures complex inter-data relationships for accurate aging assessment. A lightweight GNN model optimizes computational efficiency, ensuring effective performance on resource-limited edge devices. Compared to traditional methods, the proposed MMF-GNN model achieves a 93.6% classification accuracy, offering significant improvements in precision and reliability. An explainability module enhances decision-making by clarifying model outputs. This approach provides a robust foundation for transformer health monitoring and maintenance strategy development in power systems.