Efficient shear stress distribution detection in circular channels using Extreme Learning Machines and the M5 model tree algorithm

Abstract

With the aid of 174 laboratory data sets, Extreme Learning Machine (ELM) and decision tree (M5) models were investigated in predicting the shear stress distribution in circular channels. To evaluate the sensitivity of the input variables, 15 different input combinations were applied to each model. The calculation results show that the Re and y/P parameter values greatly affect ELM method performance, while y/P and h/D are sensitive to shear stress distribution modeling with M5. The best models among ELM and M5 were compared with an equation based on the Shannon entropy. According to the comparison results, the two proposed models outperform the Shannon entropy equation. Moreover, the ELM method’s function is superior in estimating the shear stress distribution and more adapted to experimental data with average Root Mean Square Error (RMSE) of 0.0236 compared to the M5 method with RMSE of 0.0364.

Keywords:

• Shear stress distribution
• circular channel
• extreme learning machine
• decision tree