the choice between radial basis function and feed forward neural network to predict long term tidal condition

Possessing precise water level data in any coastal area is crucial, for any coastal engineering or managements. one of the main processes responsible for a regular water level changes is tide. due to its nature, tidal prediction is relatively easily accessible. however, the precision of the results depends on the number of constituents have been considered for the prediction. the aim of this paper is to identify the most relevant tidal constituents and their relevant amplitudes for tidal prediction in the beris port, south of iran, using artificial neural network (ann). the main constituents in the area is obtained as 𝑀2, 𝐾1, 𝑆2, 𝑁2 and 𝑂1. to regenerate the tidal condition considering these constituents two ann methods has been applied including feed forward, and radial basis functions (rbf). for the training and network test tidal data of the year 2017 has been considered. for the training a variety of months and constructions has been applied. for the feed-forward the levenberg-marquardt learning method has been considered. after executing different structures in terms of the number of neurons in the hidden layer and taking into account the minimum error and run time, the network with 5 neurons in the hidden layer and two months training was qualified. for the rbf, the radius of 2.5 has been qualified. the evaluated network of both feed-forward and rbf has been employed to reproduce tidal water level of the whole year 2018, and the results were compared with both field data and those derived from harmonic analysis. it was found that the three layers feed-forward network shows the best performance in tidal prediction with the correlation coefficient of 0.85, which is followed by rbf, with the correlation coefficient of 0.81.