改进LSTM神经网络在极短期波浪时序预报中的应用  

Application of Improved LSTM Neural Network in Time-Series Prediction of Extreme Short-Term Wave

作　　者：尚凡成 李传庆 詹可[1] 朱仁传 SHANG Fancheng;LI Chuanqing;ZHAN Ke;ZHU Renchuan(State Key Laboratory of Ocean Engineering,School of Naval Architecture,Ocean and Civil Engineering,Shanghai Jiao Tong University,Shanghai 200240,China;State Key Laboratory of Navigation and Safety Technology,Shanghai Ship and Shipping Research Institute,Shanghai 200135,China)

机构地区：[1]上海交通大学海洋工程国家重点实验室,船舶海洋与建筑工程学院,上海200240 [2]上海船舶运输科学研究所航运技术与安全国家重点实验室,上海200135

出　　处：《上海交通大学学报》2023年第6期659-665,共7页Journal of Shanghai Jiaotong University

摘　　要：高效准确的极短期预报对实海况下船海结构物的施工作业安全意义重大.由于海浪的随机性,短期预报往往使用时间序列分析进行,近年来神经网络特别是长短期记忆(LSTM)神经网络在时间序列分析上预报能力强.基于此,提出一种结合生成式对抗思想的LSTM改进形式,在神经网络中嵌入频域特性等的先验知识,实现时频域信息耦合预报.经实验测试可知,该方法预报精度优于传统时序分析方法和LSTM神经网络结果,适用于极短期时序预报,有助于实现更好的船舶操纵控制.Efficient and accurate extreme short-term prediction is of great significance for the safety of ship and marine structures in actual sea waves.Due to the stochastic of actual sea waves,short-term prediction always uses time series analysis.The neural networks,particularly long short-term memory(LSTM)neural networks,have received increasing attention for their powerful forecasting capability in time series analysis.Based on this,an improved form of LSTM combining generative adversarial ideas is proposed,in which the frequency domain characteristics are embedded in the neural network to achieve coupled time-frequency domain information forecasting.The experimental test shows that the forecasting accuracy of this method is better than the results of traditional time series analysis methods and the LSTM neural network,and it is suitable for extreme short-term time series prediction for better ship maneuvering.

关 键 词：极短期预报 时序分析 长短期记忆神经网络 生成式对抗 

分 类 号：U664.82[交通运输工程—船舶及航道工程]