Intelligent Modeling for Streamflow Forecasting

In Brazil, power generation stems mostly from hydroelectric power plants and this is due to the available geographical conditions. For optimization purposes and economy of these resources, stream flows series forecast is a palpable alternative that helps in the planning of operations in hydroelectric plants. We find in the literature many models for predicting stream flow series among them, artificial neural networks, genetic programming, autoregressive models. It is proposed in this paper to build models that are able to combine several forecasters obtained by the individual components, in order to get forecasts with minor errors and therefore more accurate. For this work, we used a base National Systems Operator data (ONS) of some plants that make up the Rio Grande: Camargos, Funnel Grande, Furnas, Mascarenhas de Moraes, Jaguara, Igarapava, Volta Grande, Puerto Colombia, Marimbondo , Caconde, Euclides da Cunha Oliveira and AS. The strategy of combining forecasts obtained through different models made it possible, for the most part, the decline in the ERM, being a promising technique for streamflow forecasting series. Using the ensembles strategy was possible an improvement of up to 15,76% in the ERM.