基于TCN模型的半潜式平台运动极短期预报  

Extreme Short-Time Prediction of Semi-Submersible Platform Motion Based on TCN

作　　者：肖峰 李欣[1,2] 杨建民[1,2] 郭孝先 李琰[1,2] XIAO Feng;LI Xin;YANG Jianmin;GUO Xiaoxian;LI Yan(State Key Laboratory of Ocean Engineering,Shanghai Jiao Tong University,Shanghai 200240,China;Yazhou Bay Institute of Deepsea SCI-TECH,Shanghai Jiao Tong University,Sanya 572024,Hainan,China)

机构地区：[1]上海交通大学海洋工程国家重点实验室,上海200240 [2]上海交通大学三亚崖州湾深海科技研究院,海南三亚572024

出　　处：《海洋工程装备与技术》2022年第4期1-9,共9页Ocean Engineering Equipment and Technology

基　　金：国家自然科学基金资助项目(51779141)“海底溃坝式异重流演化机理及与海洋结构物动力响应研究”;海南省重大科技计划项目(ZDKJ2019001)“深海装备实测运维数字化系统研究与应用”。

摘　　要：半潜式平台在深海中受到环境作用会产生6个自由度运动,这种不确定的运动对平台作业造成不利影响。因此,在较短时间内准确预报平台运动响应具有工程意义。目前,基于神经网络的时间序列预报方法比较主流的模型是长短期记忆(LSTM)网络,针对LSTM网络参数多、计算量大的问题,本文提出了基于时间卷积网络(TCN)模型进行半潜式平台运动极短期预报的方法。该方法以半潜式平台模型为研究对象,将TCN应用到时间序列的预报中,选取百年一遇、千年一遇的波浪条件,针对半潜式平台的垂荡运动,进行提前量为0、5s和10s的预报,将TCN模型与LSTM网络模型的预报结果进行对比,结果表明,在达到LSTM网络模型相近预报精度的情况下,TCN模型结构更简单,故本文提出的方法是可行的。In the deep sea,the semi-submersible platform will produce six degrees of freedom motion under the action of environment,which will have an adverse impact on platform operation.Therefore,it is of engineering significance to accurately predict the motion response of the platform in a short time.At present,the popular model of the time series prediction method based on neural network is the long short-term memory(LSTM)network.Aimed at the problems of many parameters and large amount of calculation of the LSTM network,an extreme short-time prediction method of semi-submersible platform motion based on the temporal convolutional network(TCN)model is proposed in this paper,which,taking the semi-submersible platform model as the research object,applies the TCN to the prediction of time series,selects the wave conditions with a 100 year return period and a 1000 year return period,and predicts the heave motion of the semi-submersible platform with an advance of 0,5s,and 10s.A comparison of the prediction results of the TCN model and the LSTM network model shows that the structure of the TCN model is simpler but achieves the similar prediction accuracy as the LSTM network model.Therefore,the method proposed in this paper is feasible.

关 键 词：时间序列 运动预报 极短期预报 时间卷积网络 长短期记忆网络 

分 类 号：P751[交通运输工程—港口、海岸及近海工程]