Wind power prediction in Poland using temporal fusion transformers and numerical weather prediction

Abstract

Predicting wind power generation is essential to ensure the stability and efficiency of power systems. Accurate predictions enable better planning and management of energy reserves, minimizing operational costs and helping grid operators mitigate the adverse effects of wind generation fluctuations. The primary objective of the presented study is to develop an accurate wind power prediction method and apply it to Poland’s conditions. Among many emerging methods, the temporal fusion transformers (TFT) method is particularly well-suited for wind power generation forecasting, as it models complex, nonlinear dependencies in time series data. The TFT method combines self-attention mechanisms and recurrent networks, capturing long-term dependencies and short-term changes in input data. Additionally, TFT enables the effective use of contextual information, improving forecast accuracy. The numerical weather data was collected, and the feature extraction was performed. The features, such as time series data, have been used to train and test the different TFT networks. After the training and testing stage, an error analysis was performed. The final results showed similar or improved accuracy in wind generation forecasts compared to other methods in increased variability of weather conditions.