ANN-based prediction models for green water events around a FPSO in irregular waves

Abstract

The green water problem has been a critical concern for the safety of offshore structures in extreme environmental conditions. However, accurately predicting the green water events using the conventional diffraction analysis method is challenging, primarily due to the highly nonlinear physics of the green water phenomenon. In this study, a novel methodology for the prediction green water events is proposed by integrating the linear diffraction method integrated with deep learning techniques. The proposed methodology is designed as a screening tool to predict the occurrence of green water events in terms of relative wave motions around the hull. In the proposed model, a linear prediction model is employed to account for the dominant physics of ship hydrodynamics and wave mechanics, while the ANN model is introduced to complement its nonlinear effects. First, the prediction performance was compared between the linear model and the ANN-based models for the peak value of relative wave motion. The ANN-based prediction model significantly reduced the prediction error of the linear prediction model. Moreover, the ANN model based on the time series itself shows a higher predictive performance than the ANN model using features extracted from the time series. Next, the predictive performance of the ANN model was analyzed when applied to different wave conditions. It was found that the prediction error greatly decreased even for different wave conditions. Finally, the screening performances of the prediction models are investigated. The ANN-based prediction model was found to be a more effective screening tool than the linear prediction model in terms of screening quality index.