Evaluation of discharge coefficient of triangular side orifices by using regularized extreme learning machine

Abstract

The present paper attempts to reproduce the discharge coefficient (DC) of triangular side orifices by a new training approach entitled “Regularized Extreme Learning Machine (RELM).” To this end, all parameters influencing the DC of triangular side orifices are initially detected, and then six models are extended by them. For training the RELMs, about 70% of the laboratory measurements are implemented and the remaining (i.e., 30%) are utilized for testing them. In the next steps, the optimal hidden layer neurons number, the best activation function and the most accurate regularization parameter are chosen for the RELM model. As a result of a sensitivity analysis, we figure out that the most important RELM model simulates coefficient values with high exactness. The best RELM model estimates coefficients of discharge using all input factors. The efficiency of the best RELM model is compared with ELM, and it is demonstrated that the former has a lower error and better correlation with the experimental measurements. The error and uncertainty examinations are executed for the RELM and ELM models to indicate that RELM is noticeably stronger. At the final stage, an equation is proposed for computing this coefficient for triangular side orifices and a partial derivative sensitivity analysis is also carried out on it.