An attempt to predict planing hull motions using machine learning methods

Abstract Designing a high-speed craft for better seakeeping in waves can contribute significantly to higher safety and human comfort. Early the design process, mathematical models such as 2D+T method are commonly used, while high-fidelity computational fluid dynamics (CFD) experimental used later process. Some of limitations that they not fast enough be ship’s system real-time monitoring or develop digital twin. Recently, machine learning methods have demonstrated great promise building surrogate from data. These include deep recurrent neural network (RNN). In this paper, systematic investigation architectures optimizers train is presented. Adam, Adagrad, RMSprob SGD investigated training network. To model almost 35000 data points were collected Fridsma hull operating 18 regular using model. The result showed gated unit (GRU) outperformed long short-term memory (LSTM) RNN predicting heave motion. Also, one hidden layer with 5 neurons was achieve mean absolute error 0.000298 predict unseen when trained more than 24000 points.