Energy dissipation in culverts is a complex phenomenon due to the nonlinearity and uncertainties of the process. In the current study, the capability of Gaussian process regression (GPR) and support vector machine (SVM) as kernel-based approaches and the gene expression programming (GEP) method was assessed in predicting energy losses in culverts. Two types of bend loss in rectangular culverts and entrance loss in circular culverts with different inlet end treatments were considered. Various input combinations were developed and tested using experimental data. The OAT (one-at-a-time), factorial sensitivity analysis and Monte Carlo uncertainty analysis were used to select the effective parameters in modeling. The results of performance criteria proved the capability of the applied methods (i.e. high correlation coefficient (R) and determination coefficient (DC) and low root mean square error (RSME)). For rectangular culverts, the model with parameters Fr (Froude number) and θ (bend angle), and for circular culverts, the model with parameters Fr and Hw/D (depth ratio), were the superior models. It showed that using the bend downstream Froude number caused an increment in model efficiency. Among the four end inlet treatments, mitered flush to 1.5:1 fill slope inlet yielded more accurate prediction. The sensitivity and uncertainty analysis showed that θ and Hw/D had the most significant impact on modeling, and Fr had the highest uncertainty.

You do not currently have access to this content.