An Artificial Neural Network Approach for Parametric Rolling Prediction

摘要

Parametric rolling is a major issue affecting nowadays modern fishing and containership fleets. This sudden phenomenon implies huge rolling oscillations that can cause cargo damage, crew injuries and may lead to the loss of the vessel in the most severe cases. Due to its short development time and the nonlinear behaviour of parametric rolling, a real time detection system could not be fast enough to generate a response able to take the ship out of the risk area. Thus, a prediction system appears to be an adequate tool to provide information in advance, allowing the crew or a control system to take corrective actions. The present work describes the ongoing activity in a research project we are working on for the development of a parametric rolling prediction and detection system. This system is based on artificial neural networks and on their well known capabilities for predicting nonlinear system behaviour. In this case, ship motions are simulated using the three degree of freedom nonlinear coupled model developed by Neves and Rodriguez (2006), which has been proven to accurately reproduce parametric rolling. The ship used for this analysis is a transom stern fishing vessel having an identified tendency to develop parametric rolling. The Neural Networks used here take roll angle as system inputs producing a forecast of rolling angle as output. These networks are trained by using Neves and Rodriguez (2006) model as simulator in different longitudinal wave conditions leading to parametric rolling. Finally, the trained networks are used to predict ship behaviour for different time windows in advance, obtaining very promising results for making short term predictions that can allow crews to take actions and prevent the appearance of parametric rolling or even to activate an automatic system which takes these actions by itself.