Assessment of discharge coefficient in trapezoidal and rectangular canals through regularized extreme learning machine

Abstract

Since side weirs efficiently serve in flood management plans, irrigation canals and drainage systems, the estimation or simulation of its discharge coefficient seems to be a necessary task. Moreover, as the determination of this coefficient is perhaps the most important factor for the design of a side weir, there are numerous studies focused on this regard. As for the first case, this paper aims at the utilization of the modern regularized extreme learning machine (RELM) method for simulating the discharge capacity of side weirs within trapezoidal and rectangular flumes. At the first step, the parameters affecting the discharge coefficient are detected, and then 19 RELM models are extended using them. To train and test the RELM models. 70% and 30% of the experimental data are employed, respectively. In addition, the hidden layer neurons and also the activation function belonging to the RELM model are optimized. In other words, the number of neurons is considered as 14 and the activation function is introduced as the best. The superior model predicts the target values using the Froude number, the slope of main conduit walls and ratio of the weir crest to upstream flow depth. The superior model of RELM is very powerful and even shows a better performance in comparison with the ELM model. For instance, the correlation coefficient, Scatter Index and Nash-Sutcliffe Efficiency Coefficient for the RELM superior model are estimated to be 0.982, 0.043 and 0.963. A formula is suggested for approximating the target function for application works. Lately, a partial derivative sensitivity analysis (PDSA) is executed for the provided equation.