Fuzzy neural networks for river stage forecasting with uncertainty

This paper proposes a new procedure for river stage forecasting under uncertainty based on the use of artificial neural networks whose parameters are represented by fuzzy numbers. These fuzzy parameters are estimated using a calibration procedure that imposes a constraint whereby for any assigned h-level (level of credibility) the envelope of the corresponding intervals representing the river stages forecasted at different time instants must include a preselected percentage of observed values. The application of the fuzzy neural network to a real case study and the comparison of the results with those provided by Bayesian neural networks shows the validity of the proposed procedure.