Transformer Fault Prediction Method Based on Weighted K-Means and Integrated FP-Growth Algorithm

To overcome the inherent limitations of conventional transformer fault prediction methods in cluster partitioning and association rule mining, this study proposes a hybrid framework combining weighted K-means clustering with an enhanced FP-Growth algorithm. The methodology systematically quantifies the heterogeneous contributions of five critical fault indicators–dissolved gases (methane (<tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\text{CH}_{4}$</tex>), ethane <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\left(\mathrm{C}_{2} \mathrm{H}_{6}\right)$</tex>, ethylene <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\left(\mathrm{C}_{2} \mathrm{H}_{4}\right)$</tex>, total hydrocarbons) and winding temperature–through entropy-weighted feature assignment, thereby establishing an adaptive weighted K-means model for precise fault pattern classification. Furthermore, a multi-threshold FP-Growth algorithm is developed, integrating an anomaly-aware data compensation mechanism and dynamic support threshold optimization to address the underrepresentation of rare fault patterns. Case study results demonstrate that the proposed method achieves 89.0% fault matching accuracy (12.7% improvement over traditional K-means) and significantly enhances predictive performance, with ROCAUC and PR-AUC scores reaching 0.966 and 0.929, respectively (<tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{1 5. 3 \%}$</tex> and <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{2 9. 4 \%}$</tex> improvements over conventional FP-Growth). These results validate the framework's effectiveness in detecting rare fault patterns, such as insulation aging triggered by minor gas increments, with a 19.3% recall rate improvement for anomalies.