Artificial Neural Network Prediction of Overtopping Rate for Impermeable Vertical Seawalls on Coral Reefs
Ye Liu, Shaowu Li, Xin Zhao, Chuanyue Hu, Zhufeng Fan, Songgui Chen
Abstract
An artificial neural network (ANN) tool trained using a backpropagation algorithm was developed to predict the overtopping rate of impermeable vertical seawalls on coral reefs. The training database was produced from simulations of a nonhydrostatic wave model calibrated using a subset of experimental overtopping data and covered a wide range of hydrological conditions, reef morphologies, and seawall heights. The ANN configuration was optimized through sensitivity analysis and overfitting was prevented using the k-fold cross-validation technique. The generalization ability of the ANN tool was tested against the remaining subset of the experimental data. The ANN tool provided reliable predictions using deep water wave parameters as input rather than parameters for waves at the toes of structures. This made it a practical predictor for use in the preliminary design of vertical seawalls and real time forecasting of wave-induced flooding in coral reef environments.