Enhancing Wind Speed Forecasting with Multi-Meteorological Factors LSTM Models and Novel Loss Function Design

Wind power generation has grown in importance as renewable energy technology has advanced; nonetheless, wind speed forecasting remains a concern. Traditional approaches are inadequate in dealing with the nonlinearity and long-term dependency of wind speed data. The suggested method aims to present a deep learning architecture that incorporates several meteorological elements, with a hybrid loss employed to increase the wind speed prediction effect. The experiment used meteorological data from the Hangzhou Asian Games. Comparative results show that our suggested strategy outperformed ARIMA and Random Forest models in measures by up to 10.86%. Ablation studies demonstrated that including wind speed along with other meteorological data can improve performance, highlighting the importance of wind speed in predicting.