Modelagem estocástica de séries mensais apresentando dependência de longo termo para dimensionamento de reservatórios de regularização
Nenhuma Miniatura disponível
Arquivos
Data
2011-04-01
Autores
Coser, Marisa Cruz
Título da Revista
ISSN da Revista
Título de Volume
Editor
Universidade Federal do Espírito Santo
Resumo
Hydrological streamflow series analysis is fundamental for the design of reservoirs for controlling rivers water flows. This study addresses the stochastic modeling of monthly streamflow series presenting long dependency, by researching alternatives to reduce the number of parameters for models, seeking, however, to avoid losses in the preservation of statistical characteristics related to long dependence present in existing long-term historical series. There were reduced the number of complete multiplicative PARMA models parameters by subtracting those corresponding to predominantly drought and flood semesters and it was verified the behavior of these derived models with respect to historical statistical parameters and reservoirs volumes preservation. Derived models were fitted to 138 time series of monthly flows featuring long-range dependence and compared, with regard to preservation of statistical parameters and estimated volumes of reservoirs, with traditional and full multiplicative PARMA models. Comparisons were made by calculating the average percentage errors of reproduction of historical data monthly means and standard deviations, lag1 and Lag12 monthly and lag1 annual autocorrelations, Hurst coefficients and estimated reservoirs volumes. It was concluded that models with subtraction of parameters, developed in this study, are alternatives for preservation of historical long-term dependence related characteristics and volume of reservoirs if there is concern about the principle of parsimony. It was concluded, further, that the Portmanteau tests and information criteria, traditionally used for selection of models, favor those presenting lower numbers of parameters, even when they have much lower performance than those proposed in this study with respect to preservation of historical statistical features related with long memory and reservoir volumes. Multiple synthetic series generation allowed risk analysis of necessity of larger flows regularization volumes, by considering different sequences of monthly flows with equal occurrence probability.