Efficient water resource management is critical for hydropower generation and sustainable energy planning. This study investigates enhanced evolving fuzzy systems (eFS) for streamflow forecasting within the Brazilian hydrological system. Two innovative approaches are proposed by incorporating adaptive hyperparameter update mechanisms into the evolving Multivariable Gaussian (eMG) model: the Variable Step-Size Adaptive evolving Multivariable Gaussian (VSA-eMG) and the Modified Variable Step-Size evolving Multivariable Gaussian (MVS-eMG). This is the first application of these adaptive mechanisms within eFS for short-term streamflow forecasting. The models are evaluated against several benchmarks, including the operational Soil Moisture Accounting Procedure implemented by the Brazilian National System Operator (SMAP/ONS) and other data-driven evolving models. The analysis uses a nine-year dataset (2011–2019) from four strategic Brazilian basins (Furnas, Itaipu, Itumbiara, and São Simão), validated through an expanding-window scheme with seven independent 14-day folds. Results demonstrate that the proposed approaches statistically outperform non-evolving benchmarks and exhibit competitive performance compared to existing evolving models. Specifically, the MVS-eMG model achieved the best overall performance with an average Mean Absolute Percentage Error (MAPE) of 2.32% and a near-zero percent bias (0.17%). This represents an 89% reduction in forecasting error compared to the 21.29% MAPE of the operational SMAP/ONS benchmark. Statistical validation using the Morgan–Granger–Newbold (MGN) test confirmed the significance of these improvements in all tested basins. Due to their adaptive nature, the proposed models effectively handle the variability inherent in streamflow observations, demonstrating a promising balance of accuracy, computational efficiency, and adaptability for practical hydrological forecasting.
Rocha et al. (Thu,) studied this question.