New neural network-derived empirical formulas for estimating wave reflection on Jarlan-type breakwaters

A new semi-empirical model is used to estimate the coefficient of reflection for single- and double-perforated chambers in Jarlan-type breakwaters. This semi-empirical model is based on a potential flow theoretical model which was modified with specific, empirical formulas to obtain a much better agreement with the experimental tests. Single-chamber and double-chamber slotted and perforated Jarlan-type breakwaters were tested with 1500 regular wave and 160 random wave runs. Pruned Neural Network models with Evolutionary Strategies were used to identify the nonlinear relationships between the structural and wave attack parameters and the Jarlan-type breakwater reflectivity.This new semi-empirical model is valid for regular and random waves on single-chamber and double-chamber Jarlan-type breakwaters, providing estimations of the coefficient of reflection with a relative mean squared error lower than 10% for all experimental observations used to calibrate the model.