Hybrid LSTM–XGBoost Prediction of Power System Dynamic States Under Renewable Integration

With the increasing penetration of renewable energy and inverter-based resources, power systems exhibit stronger uncertainty and nonlinear dynamic characteristics, which increases the need for accurate short-term prediction of dynamic states. This study proposes a hybrid prediction method combining Long Short-Term Memory (LSTM) networks and XGBoost to improve the forecasting accuracy of key dynamic variables. The LSTM module is used to extract temporal dependencies from historical time-series data, and the extracted deep features are fused with the raw input features to construct an augmented feature vector. An XGBoost regressor is then employed to capture nonlinear feature interactions and generate the final prediction results. The proposed method is evaluated using rotor speed, active power, and power angle as representative dynamic variables. Test-set results in physical units show that the proposed model achieves RMSE values of 1.0 × 10−6 p.u., 2.5808 MW, and 0.0001 deg, and MAE values of 1.0 × 10−6, 1.0269 MW, and 0.0001 deg, respectively. Compared with the reference model, the proposed method reduces both RMSE and MAE for all three variables, indicating that the LSTM-XGBoost framework can improve dynamic-state prediction accuracy in power systems.