| dc.description.abstract | The increasing global population has raised concerns about the sustainability of energy supply. Renewable energy sources like solar and wind power have garnered significant attention due to their environmental benefits. Predicting short-term solar and wind energy production accurately is crucial for integrating these renewable sources into the electricity grid efficiently. Reliable forecasts are essential to maintain the stability and reliability of the grid. Utilizing advanced technologies such as Artificial Intelligence (AI) can significantly enhance short-term forecasting accuracy, leading to a more effective utilization of renewable energy sources. In this study, sophisticated ML algorithms as a subset of AI, including LSTM, RF, and XGBoost, are employed to predict short-term solar and wind energy generation. Each algorithm’s capabilities are harnessed to capture the temporal patterns in energy generation data, ultimately improving prediction accuracy. These three models were trained, evaluated, and compared. The best-performing model was used for prediciton purposes. Two distinct sets of data for solar and wind power in Norway and Sweden were utilized. Two distinct solar power variations were employed to maximize data utilization. The first version, a univariate dataset, exclusively comprised solar power data, utilizing only previous power data for predictions. In the univariate dataset, features encompass both solar power and meteorological variables. Regarding solar power data, the XGBoost model demonstrated reasonable performance with an 97% R2 score on the univariate dataset and a commendable 96% R2 score on the multivariate dataset. In the wind power dataset’s distinct regions (SE1-SE4), RF exhibited acceptable strength, particularly in SE1 and SE2, achieving reasonable R² values of 97%. Additionally, both XGBoost and LSTM models performed well, attaining R² scores of 96% for SE1 and 97% for SE2. Furthermore, XGBoost demonstrated outstanding performance in SE3 and SE4, securing an impressive 98% R² score. Following closely, the RF model achieved an acceptable R² score of 98% for SE3 and SE4 regions. Moreover, the research findings indicated that the precision of both univariate and multivariate solar and wind power predictions remained relatively consistent up to a forecast horizon of 7 hours. However, there is a noticeable decline in accuracy as the forecasting horizon extends to 24 hours. Moreover, the optimal prediction horizon was determined to be 24 hours ahead for SE1 region, while for regions SE2, SE3, and SE4, the ideal prediction horizon was identified as 7 hours. | |