Analysis of Wave Directional Spreading Using Neural Networks

Abstract

The short-term directional spreading of wave energy at a given location is popularly modeled with the help of the Cosine Power model. This model is oriented mainly around value of the spreading parameter involved in its expression. This paper describes how a representative spreading parameter could be arrived at from easily available wave parameters such as significant wave height and average zero-cross wave period, using the technique of neural networks. It is shown that training of the network with the help of observed directional wave (e.g., heave-pith-roll buoy) data could be used to establish dependency of the spreading parameter on more commonly available unidirectional wave parameters derived from, for example, pressure gauge data. It is found that such a procedure involving neural networks is much more accurate and reliable than the conventional approach based on statistical linear regression.