A hybrid wind speed forecasting model based on discrete wavelet transform（DWT）and bidirectional recurrent neural networks to address the prediction challenges caused by the non-stationary characteristics of wind speed data is proposed. The model employs a three-stage architecture：first, DWT decomposes the non-stationary wind speed sequences into multiple frequency sub-bands to extract multi-scale features；second, each sub-band is fed into bidirectional long short-term memory networks（BiLSTM）and bidirectional gated recurrent units（BiGRU）for parallel processing to fully capture long-term and short-term temporal dependencies；finally, a meta-learner intelligently fuses all sub-model predictions to generate the final wind speed forecast. Experiments on real data from the Sotaventogalicia wind farm in Spain demonstrate that the proposed model significantly outperforms traditional methods and existing DWT-based models across all evaluation metrics. The DM statistical test confirms the statistical significance of the performance improvement, indicating that this hybrid model provides a high-accuracy solution for wind speed forecasting.