Three Different Adaptive Neuro Fuzzy Computing Techniques for Forecasting Long-Period Daily Streamflows

A modeling study was presented here using three different adaptive neuro-fuzzy (ANFIS) approach algorithms comprising grid partitioning (ANFIS-GP), subtractive clustering (ANFIS-SC) and fuzzy C-Means clustering (ANFIS-FCM) for forecasting long period daily streamflow magnitudes. Long-period data (between 1936 and 2016) from two hydrometric stations in USA were used for training, evaluating and testing the approaches. Five different input combinations were applied based on the autoregressive analysis of the recorded streamflow data. A sensitivity analysis was also carried out to investigate the effect of different model architectures on the obtained outcomes. When using ANFIS-GP, the double-input model gives the best results for different model architectures, while the triple-input models produce the most accurate results using both ANFIS-SC and ANFIS-FCM models, which is due to increasing the model complexity for ANFIS-GP by using more input parameters. Comparing the all three algorithms it was observed that the ANFIS-FCM generally gave the most accurate results among others.