基于改进LSTM的船舶操纵运动在线预报方法研究  

Online Prediction Method of Ship Maneuvering Motion Based on Improved Long-Short Term Memory Neural Network

作　　者：董磊[1] 马翔[2] 封佳祥 孔令军 刘义 王鸿东[1] DONG Lei;MA Xiang;FENG Jiaxiang;KONG Lingjun;LIU Yi;WANG Hongdong(MOE Key Laboratory of Marine Intelligent Equipment and System,Shanghai Jiao Tong University,Shanghai 200240,China;Marine Design and Research Institute of China,Shanghai 200011,China;Science and Technology on Water Jet Propulsion Laboratory,Shanghai 200011,China)

机构地区：[1]上海交通大学海洋智能装备与系统教育部重点实验室,上海200240 [2]中国船舶及海洋工程设计研究院,上海200011 [3]喷水推进技术重点实验室,上海200011

出　　处：《中国造船》2023年第2期184-198,共15页Shipbuilding of China

基　　金：国家自然科学基金面上项目(52271348);船舶总体性能创新研究开放基金(31422116)。

摘　　要：为了提高船舶操纵性运动在线预报模型的泛化能力,针对航行船舶所处海洋环境的实时变化,将长短期记忆(LSTM)神经网络模型与注意力机制相结合,提出了改进的长短期记忆神经网络模型。基于船舶在不同波高和浪向下回转运动的仿真试验数据,设计神经网络的学习样本。通过对样本的学习,完成了船舶操纵性运动的在线预报。研究结果表明,改进的长短期记忆神经网络模型可在新的波浪环境下对船舶操纵性运动进行准确的在线预报。In order to improve generalization ability of the online model predicting ship maneuvering motion in the marine environment,an improved long and short-term memory(LSTM)neural network model is proposed by combining the LSTM neural network with the attention mechanism.Training datasets of the neural network are designed based on the data of ship's simulated turning test in waves with different wave heights and directions.Prediction of ship maneuvering motion is completed through sample studies.The results indicate that the improved long-short term memory neural network model can make accurate online prediction of ship maneuvering motion in a new wave environment.

关 键 词：船舶操纵 长短期记忆神经网络 注意力机制 

分 类 号：U661.3[交通运输工程—船舶及航道工程]