Caracterização de rede pluviométrica do Estado de Sergipe e aplicação das redes neurais para preenchimento de falhas

Abstract

The present work has as its primary objective to characterize the pluviometric system in the state of Sergipe and evaluate the applicability of the artificial neural networks in the process of fulfilling malfunctions in the historical series of pluvimetric data. It is intended, with this work, to socialize the technical knowledge about the existing system in the state and present to the users data of precipitation, a possible tool for the treatment of this information. To characterize the pluviometric system in the state of Sergipe, interviews were carried in public organizations, at the state and federal level, that have pluviometric posts or stations in Sergipe, aiming at identifying the profile of the people responsible for the stations; the structure of the network; and the process of monitoring and validating the precipitation data. To evaluate the functioning of the neural networks when the failures were fulfilled, three experiments were carried out, varying the structure of the created system. The results of the research reveal that Sergipe has a pluviometric system with good density, but poor distribution of stations. Also, a considerable number of stations today are inactive, and amongst the ones that are active, the great majority is conventional, and models that do not meet the recommendations of the World Meteorology Organization. This points out the urgent need to restructure the pluviometric system in the state. It has also been discovered that failures in the process of collection and transmission of data generate innumerous problems and inconsistence in the data of historical series of pluviometric data. About the applicability of the neural network to overcome the last problem, the experiments showed that the neural systems are able to generate satisfactory results. Nevertheless, there is some difficulty to obtain the values with precision, because the series of pluviometric data are not obtained with regularity, which implies in problems to recognize a specific pattern. Whereas researchers aim at consistent and precise data, this technique should not be disregarded, but improved and utilized after the pre-treatment of the data through simpler models.