PREDICTION OF RIVER PIPELINE SCOUR DEPTH USING MACHINE LEARNING APPROACHES

Abstract

The precise assessment of local scour depth under pipelines is a complicated occurrence, and the definitive method for its calculation remains unclear. This work tackles the issue by using computational models to accurately predict scour depth with great reliability. A support vector machine (SVM) was used to forecast pipeline scour depth, using an extensive dataset. The results were juxtaposed with typical datasets and other prediction methodologies, including regression equations and Radial Basis Function Neural Networks (RBFNN). The comparison study indicates that the SVM surpasses conventional regression techniques and RBFNN, attaining a superior generalization capability with R² = 0.89, RMSE = 0.046, MAE = 0.32%, and δ = 9.9. Principal results indicate that the mean diameter of particles substantially affects scour depth, while flow discharge has no effect. Non-dimensional metrics, like the Shields parameter, are essential in assessing scour depth. These findings underscore the efficacy of SVM in precisely forecasting scour depth under pipelines, making it an indispensable instrument for hydraulic engineering applications. This paper introduces a unique contribution to scouring estimate approaches by using both dimensional and non-dimensional datasets, establishing a baseline for future research in this field.