Predicting Hydrogen Production Using XGBOOST: a Comparative Study of Energy Conversion Scenarios

Sacramento, Diosael Ferreira, Viana, Tainan Sousa, Rocha, Paulo Alexandre Costa

ISES Solar World Congress 2025 · Fortaleza, Brazil · 2025-11-03
Published by International Solar Energy Society (ISES)
DOI: 10.18086/swc.2025.05.12

Abstract

This study evaluates the performance of the XGBoost algorithm in predicting green hydrogen production from wind energy, comparing five energy conversion scenarios with different physical and empirical models. Time series of wind speed data from the state of Ceará (2019–2021) were used, subjected to cleaning and seasonal imputation, with 80% of the data divided for training and 20% for testing. The results indicated distinct performances among the test scenarios: C1 (RMSE = 201.95 kg/km²; R² = 0.72), C2 (RMSE = 86.33 kg/km²; R² = 0.54), C3 (RMSE = 280.53 kg/km²; R² = 0.73), C4 (RMSE = 45.19 kg/km²; R² = 0.49), and C5 (RMSE = 220.74 kg/km²; R² = 0.72). Scenario C4 presented the best Skill Score (0.83) and the lowest mean error (MAE = 31.72 kg/km²), standing out for its physical accuracy and stability of predictions. C2 demonstrated the shortest processing time (649 s) and good relative accuracy (MAPE = 24.02%). The results confirm that integrating physical modeling and machine learning enhances accuracy and computational efficiency.

Keywords

Green Hydrogen Forecasting, Machine Learning, XGBoost, Wind Energy

← All proceedings papers