An AI Weather Downscaling Method for Wind Power Forecasting Considering Terrain Features

The weather prediction results are critical for wind power forecasting, even a minor deviation in wind speed prediction can result in significant power forecasting error. Numerical Weather Prediction (NWP), which is mainly used for power forecasting, the spatial resolution is low and cannot encompass each wind turbine within the wind farm, thus affecting the power forecasting accuracy. Terrian is the key factor that affects the flow field inside the wind farm. To enhance the precision and reliability of weather prediction, an AI weather downscaling method for wind power forecasting considering terrain features with global forecast field as input is proposed in this paper. Firstly, an improved U-Net-based downscaling model is developed to effectively extract and integrate the correlations among meteorological elements and their spatial correlations. Then, a terrain perception fusion model based on multi-task learning is established to extract topographic features of the wind farm, including elevation, slope, aspect, and curvature of the change of slope. Finally, these features are integrated with the aforementioned downscaling model through the attentional mechanism, thereby realizing a downscaling approach considering terrain features. A wind farm in Northern China is used as an example to verify the effectiveness and robustness of the proposed method, the results show that, the prediction error can be reduced by 0.94 m/s (32.52%) when compared to NWP, 0.26 m/s (11.76%) when compared to the downscaling model without considering terrain features.