Deterministic wave prediction for unidirectional sea-states in real-time using Artificial Neural Network

Short-term deterministic wave prediction has gained increasing interest recently. This paper presents a framework on the use of data-driven model based on Artificial Neural Network (ANN) in predicting the spatial- temporal evolution of wavefields in real-time from a given wavefield record upstream. We choose a wave environment south of Albany, Western Australia, which is known to be swell-dominated, and simulate many realisations of long-crested random waves using Higher Order Spectral Method (HOSM) for each sea-state sampled from the scatter diagrams of that particular location. Several scenarios based on the number and arrangement of wave probes providing the upstream wavefield information are considered. The performance of the ANN model, after training process, is compared with a model based on linear wave theory (LWT) and expressed in terms of normalized prediction error, taking the simulated wavefields from HOSM as the reference. We will show that the use of ANN model is promising, in that it is able to provide reasonable prediction with error within 20% over a large distance downstream.