STREAMFLOW FORECASTING FOR BASINS OF THE BRAZILIAN INTERCONNECTED SYSTEM USING AN IMPROVED EVOLVING MULTIVARIABLE GAUSSIAN MODEL

This work presents a new forecasting model applied to streamflow prediction in Brazilian watersheds, called Novel Set-Membership evolving Multivariable Gaussian (NSM-eMG). The model adjusts the learning rate based on error during the training process, aiming to improve performance and increase adaptability. To validate its effectiveness, experimental tests were conducted using real data from two watersheds relevant to Brazil’s hydroelectric operations (Furnas and Itaipu). The evaluation considered error metrics and the number of rules in the evolving models. Results were compared with the reference model used by the National Electric System Operator (ONS), the Soil Moisture Accounting Procedure (SMAP). The proposed model outperformed SMAP in all scenarios, with significant reductions in both error (~70%) and bias. These results suggest that the proposed approach is flexible and promising for streamflow forecasting in watersheds.