Analise e previsões de vasões utilizando modelos de series temporais, redes neurais e redes neurais nebulosas

AUTOR(ES)
DATA DE PUBLICAÇÃO

2000

RESUMO

Analysis and forecast of seasonal stream flow series are of utmost importance in the operation planning of water resources systems. One of the greatest difficulties in forecasting of those series is the seasonality nature of stream flow series due to wet and dry periods of the year. For a long time, the use of stochastic models, based on the c1assic Box &Jenkins methodology, were the most employed alternative to the deterministic or hydrologic models in the analysis and forecast of stream flow series. This methodology requires either some kind of data manipulation to deal with the nonstationarity or the use of periodic models. Therefore the statistical procedures, requires an arduous theoretical formulation. Artificial Neural Networks (ANN), especially multilayer perceptrons with a back-propagation algorithm, have recently been suggested for time series analysis. They have the ability to deal with nonlinear input-output relationships. Their major assets are the learning ability and generalization, association and parallel search capability. These qualities enable them to identify and to assimilate some of the features of the series as seasonality, periodicity, tendency sometimes difficult to detect under noise. The capability of complex mapping of the ANN increases with the number of layers and neurons. The use of ANN usually requires the investment of a long period of time in the modeling process, as well as a considerable amount of data. ln practice, however, the parameters usually must be quickly estimated and only a small quantity of data is available. Very often, real world data are noisy, and the collected data may contain contradictions and imperfections. Tolerance for imprecision and uncertainty is also required to achieve tractability and robustness. Fuzzy sets based data analysis models have been especially suitable for these purposes. This suggests the application of neurofuzzy network models to seasonal stream flow forecasting. These models combine the advantages of the ANN and fuzzy set based approaches in a single integrated decision-making system. Analysis and forecast of stream flows one-step-ahead and multi-step-ahead are accomplished, using time series models, neural networks, and neurofuzzy networks. Database of average monthly inflows from Brazilian hydroelectric plants located in different river basins were used. The performance of the models was compared and the results show that the models here proposed provide a better performance than the others ones considering one-step-ahead forecasting and multi-step-ahead forecasting

ASSUNTO(S)

conjuntos difusos previsão hidrologica redes neurais (computação) analise de series temporais

ACESSO AO ARTIGO

Documentos Relacionados