Hybrid Forecast-Enabled Adaptive Crowbar Coordination for LVRT Enhancement in DFIG Wind Turbines.

This study proposes a hybrid forecast-enabled adaptive crowbar coordination strategy to enhance low-voltage ride-through (LVRT) performance of doubly fed induction generator (DFIG) wind turbines. A unified electro-mechanical model in the αβ/dq frames with dual closed-loop control for rotor- and grid-side converters is built in MATLAB/Simulink (R2018b), and LVRT constraints on current safety and DC-link energy are explicitly formulated, yielding an engineering crowbar-resistance range of 0.4-0.8 p.u. On the forecasting side, a CEEMDAN-based decomposition-modeling-reconstruction pipeline is adopted: high- and mid-frequency components are predicted by a dual-stream Informer-LSTM, while low-frequency components are modeled by XGBoost. Using six months of wind-farm data, the hybrid forecaster achieves best or tied-best MSE, RMSE, MAE, and R2 compared with five representative baselines. Forecasted power, ramp rate, and residual-based uncertainty are mapped to overcurrent and DC-link overvoltage risk indices, which adapt crowbar triggering, holding, and release in coordination with converter control. In a 9 MW three-phase deep-sag scenario, the strategy confines DC-link voltage within ±3% of nominal, shortens re-synchronization from ≈0.35 s to ≈0.15 s, reduces rotor-current peaks by ≈5.1%, and raises the reactive-support peak to 1.7 Mvar, thereby improving LVRT safety margins and grid-friendliness without hardware modification.