Energies, Vol. 19, Pages 930: Short-Term Forecasting of the Total Power Generation from Wind Farms and Solar Power Plants in the National Power System Using Advanced Ensemble Machine Learning Models
Energies doi: 10.3390/en19040930
Authors:
Paweł Piotrowski
The introduction of the article presents the state of renewable energy development in Poland and statistical information on its dynamics in the context of sustainable development, highlighting both the positive aspects of this situation and the potential risks to the national power system. These risks stem from the inherent instability of renewable energy generation and the seasonal variability of production. The main part of the article provides a comprehensive statistical analysis of time series data (wind energy generation and solar energy generation) aimed at identifying the appropriate input variables for forecasting models. In addition to the two time series of electricity generation, other exogenous variables and feature engineering techniques were incorporated. In the forecasting section, short-term forecasts of energy generation in the national power system from wind farms and solar power plants were developed. The forecasts for both types of renewable energy sources (RESs) were conducted separately and then integrated into a single time series to assess which forecasting approach is more effective. A detailed analysis was carried out to determine the optimal hyperparameters for individual machine learning models. Subsequently, an ensemble model was developed, integrating multiple single models. The article concludes with final insights and practical recommendations regarding the selection of preferred models and input variables that ensure the highest forecast accuracy. Additionally, potential future developments of the models and further research in this field are discussed in the context of sustainable development.