Verifying the Use of Evolving Fuzzy Systems for Multi-Step Ahead Daily Inflow Forecasting

This study presents a prediction system based on evolving fuzzy models and a bottom-up approach for daily streamflow forecasting. Prediction models are based on adaptive Takagi-Sugeno fuzzy inference systems. These models make use of a sequential learning approach for updating their own structure and parameters over time according to changes or variations identified in the series, whereas rainfall and runoff information is processed at each time instant. Models are adjusted following a bottom-up approach, which consists of dividing the global problem into sub-problems, and each sub-problem is resolved separately. Final estimate is given by the aggregation of the parts. The proposed approach is compared to the Soil Moisture Accounting Procedure (SMAP), a hydrological model used by various hydroelectric companies of the Brazilian electrical sector. Simulation studies indicate that the evolving fuzzy system presents an adequate performance, leading to a promising alternative for daily streamflow forecasting. Indeed, results are improved when predictors are combined, primarily for a multistep ahead prediction task.