Intelligent mathematical modeling of discharge coefficient of nonlinear weirs with triangular plan

In this paper, the discharge coefficient of triangular labyrinth weir (Cd) was predicted using group of methods, including data handling (GMDH), genetic programing (GP) and multivariate adaptive regression splines (MARS) techniques. For this purpose, related dataset including parameters on Cd were collected from literature. These methods were selected since, they are classified as smart function fitting (SFF) methods. The main advantages of SFF methods compared to other artificial intelligent methods are defining the most effective parameters on output and assigning more weights to them in mathematical expression process. Results of MARS indicated that this method with fifteen basic functions could achieve good accuracy for modeling and predicting Cd (R2= 0.98 and RMSE=0.024). Results of GMDH showed that this model includes two hidden layers and that there are five and four neurons at the first and second hidden layers, receptivity. Results of GP model declared that this model with three genes have accepted performance for modeling Cd. Comparing the performance of models with each other indicated that MARS model is more accurate compared to others. Observing the performance of models in terms of DDR index demonstrated that data dispersivity of MARS model was less than others were; hence, its results are more reliable.