Computational & Technology Resources
an online resource for computational,
engineering & technology publications
PROCEEDINGS OF THE EIGHTH INTERNATIONAL CONFERENCE ON ENGINEERING COMPUTATIONAL TECHNOLOGY
Edited by: B.H.V. Topping
The Application of Neural Networks to Predict Wind Waves
C.P. Tsai1, C.C. Teng2 and C.H. Tsai1
1Department of Civil Engineering, National Chung Hsing University, Taichung, Taiwan
C.P. Tsai, C.C. Teng, C.H. Tsai, "The Application of Neural Networks to Predict Wind Waves", in B.H.V. Topping, (Editor), "Proceedings of the Eighth International Conference on Engineering Computational Technology", Civil-Comp Press, Stirlingshire, UK, Paper 68, 2012. doi:10.4203/ccp.100.68
Keywords: back-propagation neural network, wind wave, significant wave height.
The prediction of significant wave height is of great importance for the construction of ocean and coastal structures, the operation of ships, as well as the human activities in maritime area. Many forecasting methods for ocean waves have been presented in the literature, such as the traditional SMB and PNJ manual methods , the numerical model based on differential equations  and the SWAN model  etc. These methods for wave forecasting were based on the wind and wave relationships. As a result of the complexity and randomness of the wind waves, the prediction of waves based on the simplified relationship may contain substantial error. For wind-wave prediction, Deo et al.  employed an artificial neural network (ANN) to predict wave height with real-time wind speed data. The correlation coefficients of their results, however, between the observed and the predicted data were not very satisfactory, possibly as a result of the uncertainties of the wind-wave relationship.
By applying the technique of ANNs in this paper, the lag effect of energy transfer between wind and wave is considered on the basis of the wind wave theory. That is, the significant height of the wave further relates to the wind speeds involving some previous hours. As the standard process of the ANN, the present model first evaluated the interconnection weights between the waves and the corresponding wind speeds by training the past records. The selection of the neurons of input, outputs and the hidden layers is discussed in detail in the study. Using long-term wind and wave data from one of the National Data Buoy Center (NDBC) stations in the Pacific Ocean, this study examined the validity and accuracy of the ANN model. It is found that the three-hourly average significant wave height can be predicted from training the wind speed data for the previous nine hours. The results show that the neural network model can provide a short term good prediction (five days to fifteen days) of the waves using the records of the previous twenty previous days.
purchase the full-text of this paper (price £20)