Keywords
wave prediction, decision trees, regression trees, cart algorithm, neural network
Start Date
1-7-2008 12:00 AM
Abstract
One of the most important factors in design of coastal and offshore structures is significant wave height. Thus, an accurate prediction of wave height is of great importance. In this paper, an alternative approach based on regression trees was applied for prediction of significant wave height. The data set used in this study comprises of wave and wind data gathered from deep water location in Lake Michigan, from 15 September to 10 December, 2002. In this study the data set was divided into two groups. The first one that comprises of 58 days (1392 data point) wind and wave measurement was used as training data to develop the regression tree. The second one that comprises of 29 days (686 data point) wind and wave measurement was used as testing data to verify the model. Wind speeds belonging up to six previous hours were given as input variables, while the significant wave height (Hs) was the output parameter. CART algorithm was employed for building and evaluating regression trees and outputs of models with different lags were compared. Result showed that regression trees can be used successfully for prediction of Hs. In addition it was found that error statistics of the models for prediction of Hs decrease as wind speed lag increases. Finally, the results of CART-based model, was compared with artificial neural networks, Results indicated that error statistics of neural networks were marginally more accurate than regression trees.
Prediction of Significant Wave Height Based on Regression Trees
One of the most important factors in design of coastal and offshore structures is significant wave height. Thus, an accurate prediction of wave height is of great importance. In this paper, an alternative approach based on regression trees was applied for prediction of significant wave height. The data set used in this study comprises of wave and wind data gathered from deep water location in Lake Michigan, from 15 September to 10 December, 2002. In this study the data set was divided into two groups. The first one that comprises of 58 days (1392 data point) wind and wave measurement was used as training data to develop the regression tree. The second one that comprises of 29 days (686 data point) wind and wave measurement was used as testing data to verify the model. Wind speeds belonging up to six previous hours were given as input variables, while the significant wave height (Hs) was the output parameter. CART algorithm was employed for building and evaluating regression trees and outputs of models with different lags were compared. Result showed that regression trees can be used successfully for prediction of Hs. In addition it was found that error statistics of the models for prediction of Hs decrease as wind speed lag increases. Finally, the results of CART-based model, was compared with artificial neural networks, Results indicated that error statistics of neural networks were marginally more accurate than regression trees.
