Periodic ARMA models applied to weekly streamflow forecasts

Abstract

This paper presents a weekly streamflow forecasting model based on linear ARMA (p, q) models, considering both periodic and nonperiodic models. For each week, fifty possible models are automatically analyzed. The best modeling and parameter estimation are chosen based on the minimum square mean forecast error of the whole time series. The proposed model, which has been validated by the Brazilian Multi-Utility Hydrological Studies Working Group, is illustrated in case studies with several hydraulic plants of the Brazilian Southern, Southeastern, North and Northeastern systems.