In the context of the swift advancement in offshore wind power, it is crucial to have long-term, effective, and accurate wind speed predictions. Given that the wind speed at wind turbines is influenced by numerous factors, discrepancies between the numerical model’s output and actual wind speed are inevitable. In this study, we propose the use of a Convolutional Neural Network (CNN) and Gated Recurrent Unit (GRU) Hybrid Network to learn the error between the output of the Weather Research and Forecasting (WRF) and actual measurements. This approach aims to correct the day-ahead wind predictions from WRF, enabling improved accuracy in the prediction of wind speed for 144 steps in the upcoming day. The test shows that compared with the WRF output, the R 2 of Dataset1 and Dataset2 is improved by 8.23 % and 3.07 % respectively after the correction. Considering the lack of comparison of various wind speed prediction methods at present, we design predictions that simulate realistic situations, and compare WRF-CNN-GRU with time-series prediction models, including Informer, Koopa, Non-stationary Transformers, and iTransformers. Compared with the best performance model, the results from the two datasets indicate that the R 2 of WRF-CNN-GRU is superior by 28.31 % and 28.83 % in Dataset1 and Dataset2, respectively. Additionally, cumulative error trend analysis and point-by-point error analysis reveal that after 60 steps, the performance advantage of WRF-CNN-GRU becomes evident. This ability to maintain accuracy and stability over the long forecast horizon effectively meets the operational needs of day-ahead scheduling and other related tasks, confirming the model's strong suitability for offshore wind farm applications and providing a valuable reference for practical implementation. • A WRF-CNN-GRU hybrid model is proposed for day-ahead offshore wind speed forecasting. • The WRF-CNN-GRU model outperforms statistical models with 28.31 % and 28.83 % higher R2 in two datasets. • Beyond 60 steps, WRF-CNN-GRU shows evident advantages, suitable for long-term offshore wind speed prediction. • The model enhances wind power grid integration reliability and reduces wind farm operational costs.
Huo et al. (Tue,) studied this question.