摘要
在海况环境下,进行船舶运动预测时。由于惯性传感器采集系统本身的电学特性,会产生误差偏移,影响预测的准确性。针对这一问题,在常规长短期记忆网络(LSTM)的基础上,设计改良了一种二元的LSTM网络架构。在船舶运动仿真平台上进行模拟船舶升沉运动实验,并通过惯性传感系统测量仿真平台实时积分位移进行计算验证。验证统计该网络预测结果峰差值均方差0.64%,均值均方差0.42%,峰值均方差0.57%,证实该网络较常规LSTM在船舶运动预测领域具有更好的针对性和适应性,更准确的对船舶运动进行预测。
When performing ship motion predictions in a maritime environment,the acquisition system of inertial sensors produces offset errors due to its electrical property,which can seriously affect the accuracy of the general prediction method.In response to this problem,a binary long short-term memory(LSTM)network architecture is developed based on a regular LSTM neural network.In the binary LSTM network,stimulating experiments on the ship’s heave motion are carried out on the motion simulation platform.The real-time integral displacement data of the simulation platform measured by the inertial sensing system is then calculated and verified.Prediction results of this network exhibit a standard deviation of 0.64%,0.42%,and 0.57%for the peak difference,mean,and peak value,respectively.Compared with conventional LSTMs,the binary LSTM network is found to be more targeted with better adaptability in the field of ship motion prediction,as illustrated in the obtained experimental data.Further-more,this network restores the motion prediction to the actual trajectory of a ship more accurately.
作者
张博一
胡雄
唐刚
邵辰彤
ZHANG Bo-yi;HU Xiong;TANG Gang;SHAO Chen-tong(Shanghai Maritime University,Logistics Engineering College,Shanghai 202003,China)
出处
《海洋科学》
CAS
CSCD
北大核心
2021年第9期69-74,共6页
Marine Sciences
基金
上海市青年科技英才扬帆计划项目(19YF1419100,19YF1418900)。
关键词
船舶运动预测
LSTM神经网络
频域积分位移
ship-motion prediction
LSTM neural network
frequency domain integral displacement
作者简介
张博一(1992—),男,博士生,主要从事海洋工程与物流装备安全工程相关研究,E-mail:1315701145@qq.com;通信作者:胡雄(1962—),男,教授,博士生导师,E-mail:huxiong@shmtu.edu.cn。
