Time Series Analysis and Forecasting in Water Demand Man-agement

Abstract

The prediction of water reservoir levels is essential for the efficient management of water resources, especially in urban areas. This study compares the ARIMA, Decision Tree (DT), and Prophet models applied to hourly data from Bairro Alto, Curitiba, collected between 2018 and 2020. The metrics Symmetric Mean Absolute Percentage Error (SMAPE), Mean Absolute Error (MAE), and Root Relative Mean Square Error (RRMSE) were used for horizons of 1, 6, 12, and 24 hours. The DT model showed better performance in MAE and RRMSE, mainly in short-term forecasts, while the Prophet model achieved lower SMAPE values, standing out in longer forecasts. The results indicate that the proposed approach improves demand prediction accuracy, supporting scarcity prevention and the sustainable use of water.