Aplicação de técnicas de inteligência computacional no preenchimento de falhas de séries temporais climatológicas

Abstract

The climate information have extreme importance for the society and public and private institutions. Generally, the actual data associated with climatological phenomena are obtained using sensors that are subject to failures. Thus, invalid data are generated and may negatively influence the analysis. The present study aims the application of computational intelligence techniques in the development of a hybrid model composed of Artificial Neural Networks (ANN) and Genetic Algorithms (GAs), to fill the gaps of monthly variables, as evaporation, maximum temperature and relative humidity from six regions of the state of Rio de Janeiro - Brazil. The results were compared with the Average data of the regions, Multiple Linear Regression, ANNs Multilayer Perceptron (MLP) and Radial Basis Function (RBF). The performance of the proposed model was evaluated using the correlation coefficient between the estimated data and actual data, mean absolute error, root mean square error, average percentage error, concordance index and confidence index. The evaluation results showed that, in most cases, the proposed model were more accurate than the other models indices, reaching correlation value equal to 0.99, and average percentage errors between 10% and 1%. Regarding the performance of the actual filling of data gaps, the proposed model had a mean value of 100% for the filling of evaporation data, 89.2% for maximum air temperature and 75.3% for relative humidity. Therefore, it can be concluded that the model is an alternative to filling the gaps in meteorological data