Predicting daily water tank level fluctuations by using ARIMA model. A case study

Abstract

Abstract The intrinsic dynamical features of water demand highlight the need of proper operational management of tanks in water distribution networks. In addition, due to the water resource scarcity, sustainable management of urban systems is essential. For this purpose, the aid of a predictive model is crucial since it allows to give short term forecasts that can be used to predict the oscillations of relevant parameters, i.e. tanks level and/or water demand. Urban water managers can use these predictions to implement actions aimed at the optimisation of the network function. Among several modelling techniques, the univariate time series analysis is instrumental since it allows forecasting the studied parameter by using the measurements of the parameter itself. In this paper, an autoregressive integrated moving average (ARIMA) model is calibrated on water levels data, measured in an urban tank in Benevento, Campania region (Italy) and then tested on a large dataset not used to tune the parameters. The validation and forecast phases show good performances of the model on a short-term forecast horizon demonstrating the excellent potentiality of this techniques. Finally, the residuals and errors analysis complete the work suggesting possible future implementations and improvements of this technique.