The Composition of Wavelet Transforms and Neural Network in Predicting of Surface Flow in the Strait of Hormuz

In the ocean, surface flow has an important role in heat transfer and climate change. The Sea flow prediction is of great importance in oceanography. In this study, neural network and wavelet techniques were used to predict the Strait of Hormuz surface flows. The data recorded in this area from November 1992 to December 2014 with time interval of 5 days prepared from NASA and Decomposed up to 10 sub-series using wavelet mother transform such as Rbio, Coif, Bior, dmey, Db, Sym, haar and then were used as input of neural network model. By applying wavelet and neural network weighting coefficients of each of the wavelet transformations were determined. Results showed that the wavelet generated by coif (5) has the most accurate prediction. In order to evaluate the effectiveness of favorable results in the training, validation and testing, multi-layer network with a number of different neurons in the hidden layer was used. The results show that the 6 subseries wavelet d1, d2, ..., d6 with R=0.891 and RMSE =0.025 in the test is the most appropriate number to predict the surface flow in the Strait of Hormuz.