Ultra-Short-Term Wind Power Forecasting Using a Hybrid TCN–BiLSTM Model

Ultra-short-term wind power forecasting plays a crucial role in maintaining reliable and cost-effective power system operation under high penetration of renewable energy. This study proposes a hybrid deep learning model that combines a Temporal Convolutional Network (TCN) with a bidirectional Long Short-Term Memory (BiLSTM) layer to predict wind power at 20,40, and 60 minute horizons. The TCN component extracts multi-scale temporal features through causal dilated convolutions, while the BiLSTM layer refines these representations by capturing forward and backward temporal dependencies. The proposed architecture is evaluated on a real-world SCADA dataset from a 3.6 MW wind turbine and compared with several deep learning baselines, including CNN, TCN, and TCN–LSTM models. Experimental results show that the proposed TCN–BiLSTM consistently achieves the best performance across all forecasting horizons, reaching$R^{2}$scores of$0.978,0.927$, and 0.885 for 20, 40, and 60 minutes ahead, respectively. Overall, the results demonstrate that integrating convolutional temporal feature extraction with bidirectional recurrent learning improves the representation of complex wind power dynamics and provides a reliable solution for ultra-short-term forecasting in wind energy systems.