Construction of Artificial Neural Network for Alongshore Current Speed

Abstract

Available measured time-series of simultaneous waves and alongshore data sets at Anmok Beach, east coast of Korea, for two periods were chosen to construct artificial neural network. Network inputs are the wave height and the wave direction, and the output is the alongshore current speed. The first period data sets were used for training, and the second period data sets were used for testing. Three neural networks were constructed, i.e. a two-layered one (input and output layers), and two, three-layered ones (one input, one hidden, and output layer). Test results of the two networks show high correlation between predicted and measured data: the three-layered networks show lower correlation coefficient than the two-layered network. The two layers network supplies valuable coefficients which may be used for other empirical formula for the alongshore current speed. The three-layered network also shows reasonable prediction capability of the alongshore current speed when two nodes were used in the hidden layer. The three networks show superior performance on prediction of the alongshore current speeds for measured input variables.