Experimental investigation and prediction of free fall jet scouring using machine learning models

作     者：Farzin Salmasi Parveen Sihag John Abraham Meysam Nouri Farzin Salmasi;Parveen Sihag;John Abraham;Meysam Nouri

作者机构：Department of Water Engineering Faculty of Agriculture University of Tabriz Department of Civil Engineering Chandigarh University School of Engineering University of St.Thomas Minnesota Department of Water Engineering Faculty of Agriculture Urmia University Department of Civil Engineering Saeb University 

出 版 物：《International Journal of Sediment Research》 (国际泥沙研究(英文版))

年 卷 期：2023年第38卷第3期

页      面：405-420页

核心收录：

学科分类：0830[工学-环境科学与工程（可授工学、理学、农学学位）] 08[工学] 0815[工学-水利工程] 081502[工学-水力学及河流动力学] 

主　　题：Free jet Scour Gene Expression Programming (GEP) Random Forest (RF) Multivariate Adaptive Regression Spline (MARS) 

摘      要：The current study deals with the depth of scour at the location of impact between a free fall jet and a riverbed. The current study is based on extensive laboratory experiments that were designed to mimic full-scale behavior. The literature review shows that relations among hydraulic parameters for predicting the depth of scour are complex; therefore, six artificial intelligence techniques are used in the current study to capture these complex relation. A total of 120 observations are used for the analysis. Results from the experiments show that with increasing downstream water depth(h), the impinging jet causes increasingly turbulent currents and large vortices that increase the scouring of the riverbed. Increasing discharge per unit width(q) enhances the relative scour depth(D/H) while increasing the average diameter of the riverbed materials(d) decreases D/H, where D is maximum scour depth and H is the height of the falling jet. With increasing(particle Froude number Fr), the relative scour depth increases.In the current study the prediction accuracy of Gene Expression Programming(GEP), Multivariate Adaptive Regression Spline(MARS), M5P Tree, Random Forest(RF), Random Tree(RT), and Reduces Error Pruning Tree(REP Tree) techniques are evaluated using the relative scour depth(D/(H-h)). The performance evaluation indices and graphical methods suggest that the GEP based model is more accurate than other prediction methods for the relative scour depth with a coefficient of determination(R2) equal to 0.8330 and 0.8270, a mean absolute error(MAE) equal to 0.1125 and 0.0902, root mean square error(RMSE) values of 0.1463 and 0.1116, and Willmott’s Index(WI) equal to 0.8998 and 0.9014, for the training and testing stages.