Text this: Short-Term Spinning Reserve Requirement Estimation Using TimeGAN-Based Synthetic Data Augmentation and a Hybrid LSTM–XGBoost Model