Wind Power Generation Forecast by Coupling Numerical Weather Prediction Model and Gradient Boosting Machines
WRF-coupled machine learning wind power generation forecasts
- Source
- Mendeley Data
- Region
- Global
- Format
- csv
- Access
- Open access
- License
- CC BY NC 3.0
- Themes
- Renewable Energy, Machine Learning & AI, Energy Technology
Get the data from Mendeley Data
About this dataset
A low-resolution WRF model is coupled with machine learning algorithms to downscale wind power generation forecasts, using data from surrounding grid points. This hybrid model is compared to a higher-resolution WRF model in terms of accuracy and computational efficiency. The study suggests that this approach improves forecast accuracy and reduces computational costs, which is crucial for ensuring electricity supply security and avoiding penalties for wind farm owners. The results indicate that machine learning algorithms can outperform traditional downscaling methods in a shorter time frame. The study highlights the potential of coupling machine learning with numerical weather prediction models in atmospheric sciences, offering a promising avenue for future research.
How the data was collected
Raw low and high-resolution WRF model outputs for the final test day, along with the corresponding WPS and input namelists, were utilized. Turbine-based production data was used to calculate total wind farm production and determine the number of available turbines. A KML file showing the license area and coordinates of the 22 wind turbines at the Yahyalı Wind Farm was also provided. Test and training datasets, prepared for input into machine learning algorithms, were located in the "exported" subdirectory under the LowResolutionWRF directory. Results from the high-resolution WRF model were available in the "exported" subdirectory of the HighResolutionWRF directory. The final GBM model from the grid search was shared in the "r_h2o_gbm_model" directory.
For post-processing, including the construction of a gradient boosting machine model and data visualization, the R programming language was employed. Python was used for exporting atmospheric variables from the WRF model output.