Wind Power Probabilistic Forecasting Based on Transfer Learning and HHO-QRBiGRU-Attention

This study introduces a novel method for wind power generation forecasting using the Harris Hawks Optimisation algorithm to maximise data resource efficiency. The proposed HHO-QRBiGRU-attention model integrates three key components, a Bidirectional Gated Recurrent Unit network for temporal pattern recognition, quantile regression for probability estimation, and an attention mechanism to prioritise relevant features. Empirical validation using operational wind farm datasets demonstrates that the research method significantly outperforms conventional forecasting techniques in terms of accuracy and reliability, indicating its strong potential for practical applications in renewable energy forecasting.