Application of artificial neural networks to wave load prediction for coastal bridges

Document Type

Conference Proceeding

Publication Date

11-24-2017

Abstract

Predicting quickly and accurately the wave forces on the bridge superstructures from wave height and bridge structure information is important in facilitating bridge design and estimating potential damages to coastal bridges by hurricanes and tsunamis for disaster management. This study proposes a machine learning approach with Artificial Neural Networks (ANNs) as an alternative and competitive method for wave force prediction in contrast to the conventional approaches of laboratory tests and numerical simulations which are often expensive and time consuming. At first, we exploit the available Computational Fluid Dynamics (CFD) simulation models/software to generate the “ground-truth” data for ANN training. A comparison study among eight linear and nonlinear regression models including ANN and Support Vector Regression and LASSO shows that ANN outperforms all other models in terms of prediction accuracy. Various tuning experiments have been carried out in selecting an optimal network structure. We also studied the effect of input data standardization and our results indicate that input data standardization is beneficial to improve the prediction performance of ANN. Finally, the trained network structures are successfully evaluated with appreciable correlation coefficients and small root mean square errors. The results show that the proposed ANN approach is robust and capable of capturing the physical complexities for the wave force on deck prediction task.

Publication Source (Journal or Book title)

ACM International Conference Proceeding Series

First Page

526

Last Page

531

This document is currently not available here.

Share

COinS