Avaliação do desempenho de modelos par e parx na previsão de vazões para hidrelétricas da bacia do rio São Francisco

Abstract

The purpose of this paper is to evaluate the performance of simple autoregressive periodic models (PAR) and with exogenous variables (PARX) using climatic indexes with main components for the forecast of seasonal hydroelectric flows in the São Francisco river basin. The proposed models of flow forecasting use the natural flow data generated by the National Electric System Operator (ONS) and statistical techniques such as: multiple linear regression and principal components method for choice of explanatory variables. It was used 8 IRI climate indexes. The performance of the methodologies is evaluated using the NASH coefficient, mean absolute error (RMSE), bias (BIAS) and correlation. The results show that the models can represent the seasonality in the study region. The PAR presents NASH better than the PARX for one month in advance. Correlation indexes and RMSE were similar in both models. As for the BIAS, the PARX model obtained superior performance to PAR, showing that the models tend to underestimate the observed values of the plants. The inclusion of exogenous variables in the models is shown to be able to improve predictions, especially for horizons greater than 1 month.