Abstract
The pursuit of carbon neutrality is reshaping global energy systems, making the transition to renewable energy critical for mitigating climate change. However, unstable weather conditions continue to challenge energy consumption stability and grid reliability. This study investigates the effectiveness of various machine learning (ML) models at predicting energy consumption differences and employs the SHapley Additive Explanations (SHAP) interpretability tool to quantify the influence of key weather variables, using five years of data (2017–2022) and 196,776 observations collected across Europe. The dataset consists of hourly weather and energy consumption records, and key variables such as Global Horizontal Irradiance (GHI), sunlight duration, day length, cloud cover, and humidity are identified as critical predictors. The results demonstrate that the Random Forest (RF) model achieves the highest accuracy and stability (R2 = 0.92, RMSE = 360.17, MAE = 208.84), outperforming other models in predicting energy consumption differences. Through SHAP analysis, this study demonstrates the profound influence of GHI, which exhibits a correlation coefficient of 0.88 with energy consumption variance. Incorporating advanced data preprocessing and predictor selection techniques remains the RMSE of RF but reduces the RMSE by approximately 25% for the XGBoost model, underlining the importance of selecting appropriate input variables. Hyperparameter tuning further enhances model performance, particularly for less robust algorithms prone to overfitting. The study reveals the complex seasonal and regional effects of weather conditions on energy demands. These findings underscore the effectiveness of ML models at addressing the challenges of complex energy systems and provide valuable insights for policymakers and practitioners to optimize energy management strategies, integrate renewable energy sources, and achieve sustainable development objectives.
Original language | English |
---|---|
Article number | 87 |
Number of pages | 25 |
Journal | Sustainability |
Volume | 17 |
Issue number | 1 |
Early online date | 26 Dec 2024 |
DOIs | |
Publication status | Published - Jan 2025 |
Keywords
- weather variability
- renewable energy
- machine learning tools