Enhancing Fault Prediction of Ultra-High Voltage Transformers Based on a Multi-Model Adaptive Weighting Ensemble Approach for Dissolved Gas Analysis

Ultra-high voltage (UHV) transformers are critical components in power systems, where their operational reliability is essential for ensuring grid stability. Dissolved Gas Analysis (DGA) is a widely used technique for the early detection of transformer faults. However, traditional DGA-based prediction methods often struggle with the complexities and dynamics of operational environments. This study introduces a novel fault prediction approach utilizing a multi-model adaptive weighting ensemble. This ensemble integrates Long Short-Term Memory (LSTM) networks, multivariate grey prediction models, exponential smoothing, and big data analytics to cater to the specific demands of various fault types. By dynamically adjusting the weights of each model, the proposed ensemble enhances both prediction accuracy and robustness. Experimental results demonstrate that this method significantly improves fault prediction precision across diverse fault scenarios, offering a valuable tool for the maintenance and reliability of UHV transformers.