Next-day average wind speed prediction in Taiwan: A comparison of neural network and hybrid wavelet-neural network models

• Hybrid Wavelet+NARX model consistently outperforms NARX in 24-hour wind speed prediction accuracy. • Evaluation metrics include RMSE, MAE, MAPE, MSE, and R 2 hybrid model shows superior performance across all. • Study guides optimal NARX setup: selecting neuron count, delay steps, and exogenous input configuration. • Offers clear criteria for choosing wavelet families and thresholding techniques for hybrid model tuning. • Readers gain actionable insights into model selection and parameter optimization for wind forecasting tasks. One growing renewable energy resource that is crucial to the shift to a more sustainable energy system is wind energy. One of the most significant issues affecting wind energy is variation in its production. Planning and operating a wind energy station requires the use of wind speed prediction techniques, which are challenging due to the dynamic nature of wind and the impact of regional variables. Nevertheless, current prediction techniques face substantial difficulties in achieving long-term nonlinear prediction accuracy due to the complexity of wind speed data, resulting in a deficiency of wind energy projections that may lead to erroneous energy distributions. To address the prediction issues, this research proposes a hybrid approach for average wind speed models that combines the Wavelet Transform (WT) and Neural Network (NN) techniques. The next-day forecast of time series data in Taiwan is assessed in this research using an uncertainty metric related to average wind speeds. Additionally, this study compares the deep learning-based Gated Recurrent Unit (GRU), Long Short-Term Memory (LSTM), and Seasonal Auto-Regression Integrated Moving Average (SARIMA) approaches. A hybrid strategy that combines WT with LSTM and WT with GRU. The WT-NN model outperforms the other models according to the results of the suggested strategy. The effectiveness of the suggested WT-NN is assessed in comparison to the different models. These findings demonstrate the effectiveness of WT-NN in enhancing the accuracy of wind speed prediction. The study's suggested approach may help predict wind speed and wind energy production.