A robust forecasting-aided state estimator for active distribution network based on improved SMOTE-transformer2

To address key challenges in active distribution networks (ADNs), including small-sample imbalance caused by distributed generator (DG) volatility, poor dynamic adaptability of traditional prediction models, and limited state estimation accuracy, this study proposes a robust forecasting-aided state estimation (FASE) algorithm named ISMOTE-Transformer 2 -EnKF, integrating improved SMOTE (ISMOTE), Transformer 2 , and Ensemble Kalman Filter (EnKF). The algorithm adopts multi-level preprocessing (RBT-LOF outlier detection, Z-Score standardization, Random Forest feature ranking) to suppress noise; uses ISMOTE (Borderline-SMOTE and CMS-ENN) to optimize data distribution and resolve decision boundary confusion of traditional SMOTE; enhances small-sample robustness and dynamic adaptability via Transformer2 with Singular Value Fine-tuning (SVF) and two-pass adaptive inference; and achieves high-precision estimation by fusing Transformer2′s predictions with measurements through EnKF. Validated on the DTU 7 k 47-bus system across five scenarios (quasi-steady state, DG fluctuations, sudden loads, bad data and topology changes), the algorithm achieves a minimum voltage magnitude (VM) RMSE of 3.5377 × 10 -4 p.u., voltage angle (VA) RMSE of 9.7174 × 10 -3 °, and accuracy /recall > 97% in small-sample imbalance scenarios, demonstrating superior precision, robustness and adaptability.