Prediction of submarine pipeline equilibrium scour depth based on machine learning applications considering the flow incident angle

Abstract

Scouring below submarine pipelines, due to fluid, structure and soil material interaction, is a complex phenomenon involving numerous effective parameters. In this research, machine learning methods, such as the GA-BP (Genetic Algorithm based Back Propagation) neural network, RBF (Radial Basis Function) and SVM (Support Vector Machine) are presented and prediction models are built for forecasting pipeline equilibrium scour depth. The results of the prediction models are compared with observed data, which shows that the GA-BP model provides the best predictive performance for scour depth exhibiting highest correlation coefficient and lowest Root Mean Square Error as compared with RBF, SVM in live-bed conditions. Results of the sensitivity analysis indicate that Froude number (Fr) is the most effective parameter for predicting the scouring depth below the pipeline. With the increase of flow incident angle, its influence on the predicted scour depth results becomes more obvious.