Application of clustering technique to improve prediction of wind-induced (Case Study: Gulf of Mexico)

In this study, wave height for four buoyes in the Gulf of Mexico predicted using wind and wave characteristics such as mean wind speed, wind gust, wind pressure, air temperature and water temperature difference, duration and fetch length. For this purpose, the combination of kmeans algorithm and the neural network has been used. Initially, the data were sorted to eliminate the effect of the seasonal process in calculating the wave height, and then the meteorological data clustered using the kmeans algorithm from k=1 to k=10. The specified clusters were introduced as inputs of the MLP model and the mean RMSE index for k cluster was calculated. The results showed that in most buoys, the optimum number of clusters is between 8 and 10. In addition, Comparison of the results of this study with other studies showed that in previous studies, the prediction was based on the inputs of the recorded wave height of adjacent buoys or wave time delay has been applied. While in this research, only meteorological parameters have been used as inputs to increase the accuracy of prediction, which has a fairly good accuracy. Therefore, using the clustering technique increases the accuracy of the estimation at wave height.