تحليل عدم قطعيت مدل هاي شبكه عصبي و نروفازي در پيش بيني جريان رودخانه

Uncertainty Analysis of Artificial Neural Networks and Neuro-Fuzzy Models in River Flow Forecasting

River flow forecasting in water resources management is of great importance. But, due to the high uncertainty in the factors affecting the rainfall-runoff process, the results are usually problematic. One of the procedures that can alleviate this problem is incorporating uncertainty analysis in forecasted results. Such an analysis has been traditionally used for statistical methods but less attention has been given to the Artificial Neural Networks (ANNs) and the Neuro-Fuzzy (ANFIS) models. These models have gained much popularity in recent years. This research has aimed to analyze the uncertainty of these techniques for 1 to 3 months forecasting of river flow. Sofy-Chay River at Tazekand gauging station in the northwest of Iran is selected as the study site to explore the methodology. The results show that ANFIS overall gave more accurate forecasts and less uncertainty. But, when it comes to high flows, the confidence interval for the two models increases quite obviously and this increases the risk for application of the results.