By improving the forecasting accuracy of model products through wind speed deviation correction methods, reliable data support is provided for wind power generation forecasting in wind farms. Taking the Lingchuan County Wind Farm in Shanxi Province as an example, three machine learning methods recurrent neural network （RNN）, nonlinear model（NLinear）, and Transformer were employed to establish wind speed deviation correction models for CMA-WSP2.0 model products. The results show that the results of three methods are better than that of the original model products, and Transformer and NLinear perform better in improving the accuracy of wind speed at heights of 10 m meters and 100 m meters, respectively. Therefore, machine learning methods can effectively improve the quality and reliability of wind speed data, offering a more accurate data foundation for wind energy resource development and utilization, power forecasting and other fields.

machine learning  /  wind speed  /  bias correction  /  numerical weather prediction  /  wind farm  /  deep learning