Estimation of Scour Depth around Submerged Weirs Using the Novel Approach Extreme Learning Machine

Scour in vicinity of hydraulic structures is considered as one of the most important parameters to design the structures. In this study, scour pattern at downstream of submerged weirs was predicted using the novel method “Extreme Learning Machine”. In current study, in order to survey the accuracy of the numerical model, the Monte Carlo Simulations(MCs) was employed. In addition, the k-fold cross validation was utilized so as to validate the results of numerical models. Then, regarding with input parameters, five ELM models were developed. Firstly, the number of optimized hidden neurons for soft computing model was calculated. Next, the best activation function for numerical model was chosen. Analysis of activation function showed that the sigmoid activation function predicted the scour around submerged weirs with more accuracy. Additionally, results of sensitivity analysis proved that the superior model estimated the scour in terms of U0/Uc, z/ht,d50/ht. This model simulated the scour around submerged weirs with reasonable accuracy, for instance, determination coefficient and scatter index were computed 0.880 and 0.127, respectively. Moreover, the velocity parameter U0/Uc was identified as the most effective input variable. Finally, a matrix was provided to estimate the scour depth for engineers without previous knowledge of Extreme Learning Machine.