Evaluation of Statistical, Empirical, Neural Networks and Neural - Fuzzy Techniques for Estimation of Spillway Aerators

One way to decreases the damage caused by cavitation in spillways is aeration flow using aerators. The required air flow of aerator is one of the most important factors in their design. In this study, to estimate the required air flow of spillway aerators four methods were applied including of stepwise regression, Pfister empirical method, neural network (based on Levenberg- Marquardt algorithm) and the combination of fuzzy-neural (ANFIS). In order to perform of modeling, 914 experimental data on physical model of Clyde Dam spillway and 12 data of Azad Dam related to conducted tests by Water Research Center on Azad dam hydraulic model were gathered. However, the performance and error of these methods were investigated after calculating the required air flow of aerators. The results showed that the combination of fuzzy-neural has the best performance with a root mean square error (RMSE) and correlation coefficient (R) about 0.0194 and 0.968, respectively. In addition, artificial neural network, stepwise regression and Pfister empirical methods had a root mean square error equal to 0.0538, 0.0596 and 1.98, respectively.