Modelo computacional preditivo de regressão supervisionado guiado por séries temporais para otimização da cloração e fluoretação em estação de tratamento de água

Abstract

This study proposes a predictive model to optimize chlorination and fluoridation processes in water treatment plants by applying machine learning algorithms such as multilayer artificial neural networks, random forests, and XGBoost. The data analyzed, obtained from a plant with surface water intake, undergo pre-processing techniques and hyperparameter optimization to ensure accurate predictions. The findings indicate significant performance improvements after the optimization process, with reduced errors and improved generalization capacity of the models. The optimized models present RMSE ≈ 0.40 for chlorine dosage, with a coefficient of determination R² ≈ 0.90, and RMSE ≈ 0.15 and R² ≈ 0.95 for fluoridation. These observations suggest that the integration of artificial intelligence in drink water treatment operations allows efficient and automated monitoring while ensuring compliance with regulatory drinking water standards. The results highlight the potential of the approach to contribute to the sustainability of the sector by reducing chemical inputs and promoting more precise and safer operations.