Neural networks for optimization of an early warning system for moored ships in harbours

Abstract

Within the BlueSafePort project, an early warning system (EWS) for moored ships in the port of Sines as developed. In order to improve the reliability and accuracy of the system, two neural networks (NN) were trained,using wave buoys measured datasets. Numerical models results for the wave propagation are djusted and forecasts are improved.The trained neural networks were able to produce more accurate stimates for the significant wave height and mean wave period, at the buoy location, deployed in front of the ines Port. The use of the new NN led to an overall reduction of the root mean square error of around 80% compared with SWAN numerical model simulations, thus reducing potential errors in subsequential alculations and alert levels issued by the system for the moored ships.