Calibração automática de modelos chuva-vazão usando um algoritmo genético.

Abstract

In this study, the SCE-UA algorithm for optimization of model parameters is presented, which is based on the use of multiple initial complexes and the natural evolution process. This algorithm combines the strength of the Nelder & Mead simplex method (1965) with the concepts of controlled random search (Price, 1987), genetic operators extracted from nature (Holland, 1975), and mixture of complexes (Duan et al., 1992). The SCEUA algorithm was used to calibrate three rain-flow models, with different structures, in two hydrographic basins in the Northeast of Brazil: one located in the timid region of the State of Paraíba and the other in the semi-arid region of the State of Pernambuco. The results obtained indicate that this algorithm is effective and efficient in locating the global optimum values ​​of the model parameters. At characteristics of the algorithm structure and the success obtained in its application for the three selected models (Tank-Model, SMAP and SWM) show its ability to overcome the difficulties usually encountered during automatic calibration such as discontinuity in the response surface, presence of great locations, extensive valleys and non-linearity of the models. Two objective functions with probabilistic components based on the maximum likelihood theory were employed in the evaluation of the algorithm. Also presented in this dissertation are specific configurations of the Tank-Model for daily and monthly intervals with application to humid and semi-arid regions.