This study aims to improve the accuracy of wind power generation forecasting by selecting the potential locations for weather stations, which serve as crucial data sources for wind predictions. The proposed method is based on using Shapley values. First, they are assigned to stations that are already available in the region based on their contribution to forecasting error. Second, the values are interpolated to cover the area of interest. We test the hypothesis that taking weather measurements in areas with negative Shapley values leads to a decrease in the error of forecasting the volume of wind power generation. We estimate the method's impact on forecasting error by using long short-term memory neural network and linear regression with quadratic penalization. The results of this proof-of-concept study indicate that it is possible to improve the short-term wind power forecasts using additional weather observations in the selected regions. The future research should be dedicated to the expansion of the case study area to other locations, including offshore power plants.

KLYAGINA Olga;  POZO CAMARA David;  BESSA Ricardo Jorge; 

2024-12-17

Institute of Electrical and Electronics Engineers

JRC138494

979-8-3503-5518-5 (online),    979-8-3503-5519-2,   

2994-9467 (online),    2994-9440,   

https://ieeexplore.ieee.org/document/10751563,    https://publications.jrc.ec.europa.eu/repository/handle/JRC138494,   

10.1109/EEEIC/ICPSEurope61470.2024.10751563 (online),