Wave Forecasting Using Neuro Wavelet Technique

Abstract

In the present work a hybrid Neuro-Wavelet Technique is used for forecasting waves up to 6 hr, 12 hr, 18 hr and 24 hr in advance using hourly measured significant wave heights at an NDBC station 41004 near the east coast of USA. The NW Technique is employed by combining two methods, Discrete Wavelet Transform and Artificial Neural Networks. The hourly data of previously measured significant wave heights spanning over 2 years from 2010 and 2011 is used to calibrate and test the models. The discrete wavelet transform of NWT analyzes frequency of signal with respect to time at different scales. It decomposes time series into low (approximate) and high (detail) frequency components. The decomposition of approximate can be carried out up to desired multiple levels in order to provide more detail and approximate components which provides relatively smooth varying amplitude series. The neural network is trained with decorrelated approximate and detail wavelet coefficients. The outputs of networks during testing are reconstructed back using inverse DWT. The results were judged by drawing the wave plots, scatter plots and other error measures. The developed models show reasonable accuracy in prediction of significant wave heights from 6 to 24 hours. To compare the results traditional ANN models were also developed at the same location using the same data and for same time interval.