Accurate prediction of roll motion in high sea state is significant for the operability, safety and survivability of marine vehicles. This paper presents a novel data-driven methodology for achieving the multi-step prediction of ship roll motion in high sea states. A hybrid neural network, named ConvLSTMPNet, is proposed to execute long short-term memory (LSTM) and one-dimensional convolutional neural networks (CNN) in parallel to extract time-dependent and spatio-temporal information from multidimensional inputs. Taken KCS as the study object, the numerical solution of computational fluid dynamics method is utilized to generate the ship motion data in sea state 7 with different wave directions. An in-depth comparative study on the selection of feature space is conducted, considering the effects of time history of motion states and wave height. The comparison results demonstrate the superiority of selecting both motion states and wave heights as the feature space for multi-step prediction. In addition, the results demonstrate that ConvLSTMNet achieves more accurate than LSTM and CNN methods in multi-step prediction of roll motion, validating the efficiency of the proposed method.

翻译：对高海状态下滚动的准确预测对于海洋车辆的可操作性、安全性和可存活性意义重大。本文件介绍了一种由数据驱动的新方法,以在公海国家实现船舶滚动的多步预测。一个名为ConvLSTMPNet的混合神经网络,建议执行长期短期内存(LSTM)和一维共变神经网络(CNN),同时从多层面投入中提取时间和时空信息。将KOCS作为研究对象,利用计算流动动态方法的数字解决方案生成具有不同波向的船舶运动7号海中的数据。对地物空间的选择进行了深入的比较研究,同时考虑到运动状态和波高的时间历史的影响。比较结果表明,选择运动状态和波高作为多步预测的地貌空间,其优势高于ConLSTMNet在多步移动预测中比LSTM和CNN方法的准确度,从而验证了拟议方法的效率。