Power Transformer Remaining Useful Life Prediction Method Based on Dual Imbalanced Data

Remaining useful life (RUL) prediction for power transformers plays a key role in safeguarding the stability and security of power systems. However, due to the dual imbalance in both fault categories and fault severity within transformer monitoring data, it becomes challenging for predictive models to effectively extract key features from minority class samples, thereby limiting prediction accuracy. In order to solve this problem, a novel approach for RUL prediction is proposed, combining oversampling and reweighting techniques. First, a three-step data oversampling strategy is developed to balance the fault category distribution. Next, the Adaptive Boosting (Adaboost) regression method is employed to assign higher weights to samples with key features that are difficult for the model to learn, thereby enhancing the model’s predictive performance for minority classes. Finally, the proposed method's effectiveness is validated using real-world transformer data.