Short-Term Probabilistic Forecasting of Water Demand Using GPR: A Case Study in Southern Italy

Short-term water demand forecasting is a key issue for the management of smart water networks, particularly in the context of remote control and active regulation. This study analyses a real-world dataset of water demand coefficients, collected at 15 min intervals, from a municipality in Southern Italy serving approximately 73,000 inhabitants. The proposed model, based on Gaussian Process Regression (GPR) with a Rational Quadratic kernel (RQ), is compared with a statistical benchmark constructed using average patterns for each time slot by the application of the Gauss Distribution. The results show a reduction in RMSE and MAE and a better ability to track the daily dynamics of demand using the GPR approach.