Comparison between the fuzzy regression and artificial neural network method in the analysis of flow hydraulic through rockfill dams

One of the cheap methods for flood control is to use the rockfill dam withoutimpermeable layer. These dams are similar to detention dams so that the peak of theiroutput hydrograph is lower than that of input hydrograph. In order to use the rockfilldams for flood control, it is necessary to have knowledge about hydraulics of flow regimeto be able to design these dams accurately. As this type of dam consists of coarseparticles, the flow will be turbulent and hence Darcy's law and other relevant equationsfor unsaturated zones are not valid. Hydraulic analysis of flow through rockfill dam wasperformed using the fuzzy regression and artificial neural networks methods.Probabilistic regression method as a type of the fuzzy regression was used in thisresearch. The coefficients and output of model were fuzzy numbers and based onminimizing the sums of the widths of membership functions of the fuzzy regressionequation coefficients. In both methods, dam specifications as well as downstream riveror channel specifications and flow rate were the input (independent) variables whereasupstream flow rate was the dependent output variable. In artificial neural networksmethod, a multiple layer perception network was used. The neurons in both input andoutput layers were input and output data respectively. This network with three hiddenlayers gained the best results. The amounts of errors were determined using real data foreach method and were 2.3 and 0.9 cm for fuzzy regression and artificial neural networksmethods respectively.