Predicting wave forces on coastal bridges using genetic algorithm enhanced ensemble learning framework

Abstract

As an essential part of the nearshore transportation system, the safety of coastal bridges is often threatened by the hostile natural environment. To facilitate the associated risk assessment and proposition of mitigation measures for coastal bridges, it is imperative to develop an efficient and reliable model for wave forces prediction. With the advent and popularity of machine learning (ML) techniques, the use of advanced ML models has shown great potential for applications in various engineering disciplines. In this study, a genetic algorithm enhanced ensemble learning framework (GA-ELF) is proposed to facilitate efficient and reliable prediction of wave forces on coastal bridge decks. Specifically, two ensemble learning techniques, i.e., Bagging and Adaboost, are employed to establish predictive models. Four typical ML algorithms, i.e., decision tree (DT), support vector regression (SVR), K-nearest neighbor (KNN), and artificial neural networks (ANN), are chosen as the individual weak learners. Two ocean engineering cases with the datasets respectively obtained from computational fluid dynamics (CFD) simulations and flume experiments are used to investigate the effectiveness and applicability of the GA-ELF for wave forces prediction. The results demonstrate that: enhanced performance in predictive accuracy and generalization ability is achieved through the presented GA-ELF technique; the ensembled predictive models, especially the Adaboost-enhanced models can be used as efficient and reliable alternatives to classical approaches for predicting wave forces on coastal bridges. It is envisioned that the GA-ELF could be extended to predict forces on other coastal engineering structures, thus facilitating the structural design and risk assessment during their service life.