Artificial Neural Network Prediction Of Overtopping Discharges Over Vertical Structures On Coral Reef

In recent years,a large number of projects have been constructed on coral reefs in the South China Sea,and the problem of overtopping over vertical structures on coral reef topography attracts more and more attention.Because of the particularity of coral reef landform,the existing calculation methods of mean overtopping discharge over vertical structures are not applicable and the wave parameters at the toes of structures are difficult to obtain.Therefore,a new calculation method of overtopping is needed.Based on the results of physical experiments,the non-hydrostatic wave model SWASH on coral reef topography is validated.The results show that the simulation accuracy of wave parameters on reef flat can be optimized by tuning the value of wave breaking parameter ?,and more accurate simulation results of overtopping can be obtained.With the change of dimensionless relative freeboard of the structures above the design water level,the relative incident wave height and wavelength,there is no obvious change in the value of ?.When the average value of ? is 2,the model still has high accuracy and small dispersion.In this study,six input parameters of SWASH model,including slope of fore reef,width of reef flat,height of structures,water depth on reef flat,effective wave height and peak period of wave spectrum of incident wave,are analyzed and selected by referring to the design wave parameters of-20 m water depth in the South China Sea.The calibration of incident wave parameters is also made for all combinations of wave parameters.2016 numerical simulations using the SWASH model are carried out,and a Back Propagation(BP)neural network model is developed based on the training data from the numerical results.The model predicts the mean overtopping discharge over vertical structures on coral reefs by inputting the topographic parameters of coral reefs and the wave parameters in deep-water.In this study,the key parameters of the neural network model are analyzed and optimized,and the accuracy of the neural network model for predicting overtopping is verified and compared.The results show that the BP neural network model developed in this study can be used as a tool for predicting the overtopping over vertical structures on coral reef topography,and the prediction accuracy is higher than the neural network tool for predicting the overtopping over vertical structures using wave parameters at the toes of structures.