Estimating the shear stress distribution in circular channels based on the randomized neural network technique

Predicting the shear stress distribution in channels is crucial in hydraulic engineering problems. Since the equations presented by other researchers for estimating the shear stress distribution in circular channels are complicated, this study focuses on applying Randomized Neural Networks (RNN) to present easier equations for computing the shear stress distribution. The specific aim of this work is to obtain accurate shear stress distribution estimation, something proven difficult through experimental and analytical methods 176 data for four circular channel flow depths serve as the entire dataset, and half are used as the testing dataset. Sensitivity analysis is applied and 15 RNN models with different input combinations are investigated. The model with Re and yll) as input variables produces the most appropriate results in predicting shear stress distribution. The best RNN model (model 10) is also compared with an equation based on the Shannon entropy. The study provides evidence that RNN Model 10 (with average RMSE of 0.0544 and MAE of 0.0463) is capable of modelling the shear stress distribution in circular channels and is more accurate than the Shannon entropy-based equation (with average RMSE of 0.1050 and MAE of 0.0840). (C) 2017 Elsevier B.V. All rights reserved.