Comparison of Reconstruction Methods for Water Supply Systems Flow Rate Time Series

Abstract

The purpose of this paper is to compare the performance of five univariate models for the reconstruction of flow rate time series. Errors in the measurements may occur due to problems in the sensor or in the communication system with data logger, thus generating missing data in the flow rate time series. The presence of missing values in flow rate data restricts its use in network operation processes. The performance of seasonal ARIMA, Standard and double seasonality Holt-Winters, and original and improved Quevedo approach are assessed. The analysis is made considering a real Portuguese case study and 1-month of flow rate data at 1-h and 10-min period. The holidays compared to the weekdays show great differences in consumption patterns. For this reason, the effect of forecasting holidays is assessed. Obtained results evidence that the improved Quevedo model can cope with different time step intervals and type of day being forecasted, with a reduced computation time.