Feature-Weighted MMD-CORAL for Domain Adaptation in Power Transformer Fault Diagnosis

TL;DR

This work tackles distribution shifts in power transformer fault diagnosis caused by varying transformer types and operating conditions by introducing a feature-weighted domain adaptation method (MCW) that fuses MMD and CORAL with KS-derived per-feature weights. The approach encodes DGA-derived features as 2D Gramian Angular Field images and uses a CNN with a KS-weighted domain loss, where the total objective is $\mathcal{L}_{total} = \alpha \mathcal{L}_{classification} + (1-\alpha) (\beta \mathcal{L}_{MMD} + (1-\beta) \mathcal{L}_{CORAL})$, and the weighted MMD and covariance terms are calculated using a weight matrix $\mathbf{W}$. Experimental results on cross-domain datasets show MCW achieving up to 7.9% higher accuracy than Fine-Tuning and 2.2% higher than MMD-CORAL, with strong robustness when target data are scarce (e.g., 85.9% accuracy with 30% of target data). These findings demonstrate the practical potential of KS-informed feature weighting to enhance domain generalization in transformer fault diagnosis.

Abstract

Ensuring the reliable operation of power transformers is critical to grid stability. Dissolved Gas Analysis (DGA) is widely used for fault diagnosis, but traditional methods rely on heuristic rules, which may lead to inconsistent results. Machine learning (ML)-based approaches have improved diagnostic accuracy; however, power transformers operate under varying conditions, and differences in transformer type, environmental factors, and operational settings create distribution shifts in diagnostic data. Consequently, direct model transfer between transformers often fails, making techniques for domain adaptation a necessity. To tackle this issue, this work proposes a feature-weighted domain adaptation technique that combines Maximum Mean Discrepancy (MMD) and Correlation Alignment (CORAL) with feature-specific weighting (MCW). Kolmogorov-Smirnov (K-S) statistics are used to assign adaptable weights, prioritizing features with larger distributional discrepancies and thereby improving source and target domain alignment. Experimental evaluations on datasets for power transformers demonstrate the effectiveness of the proposed method, which achieves a 7.9% improvement over Fine-Tuning and a 2.2% improvement over MMD-CORAL (MC). Furthermore, it outperforms both techniques across various training sample sizes, confirming its robustness for domain adaptation.

Figures (4)

• Figure 1: Augmented Images
• Figure 2: Average Pixel Intensity Distribution Comparison for GAF Images of Source and Target Datasets.
• Figure 3: Accuracy (%) comparison for different model architectures.
• Figure 4: Accuracy (%) comparison for different values of $\alpha$ and $\beta$.