Bridge afflux estimation using artificial intelligence systems

Abstract

Most of the methods developed for the prediction of bridge afflux are generally based on either energy or momentum equations. Recent studies have shown that the energy method, which is one of the four bridge subroutines within the commonly used program HEC-RAS for computing water surface profiles along rivers, produced more accurate results than three other methods (momentum, WSPRO and Yarnell's methods) when applied to bridge afflux data obtained from experiments conducted in a two-stage channel. This work developed three artificial intelligence models (the radial basis neural network, the multi-layer perceptron and the adaptive neuro-fuzzy inference system) as alternatives to the energy method. Multiple linear and multiple non-linear regression models were also used in the analysis. The results showed that the performance of the adaptive neuro fuzzy inference system in predicting bridge afflux was superior to the other models.