数据驱动的载人舱球壳下潜过程推演与异常诊断  

Data-driven algorithm for diving process modelling and anomaly detection of deep-sea pressurized spherical shell

作　　者：姚骥 汪雪良[1,2,3] 叶聪 顾学康[1,2] 陈浩政 王雷[1,2,3] 张铮铮 YAO Ji;WANG Xue-liang;YE Cong;GU Xue-kang;CHEN Hao-zheng;WANG Lei;ZHANG Zheng-zheng(China Ship Scientific Research Center,Wuxi 214082,China;Taihu Laboratory of Deepsea Technological Science,Wuxi 214082,China;State Key Laboratory of Deep-sea Manned Vehicles,Wuxi 214082,China)

机构地区：[1]中国船舶科学研究中心,江苏无锡214082 [2]深海技术科学太湖实验室,江苏无锡214082 [3]深海载人装备全国重点实验室,江苏无锡214082

出　　处：《船舶力学》2024年第11期1710-1720,共11页Journal of Ship Mechanics

基　　金：国家重点研发计划项目(2021YFC2802300);江苏省卓越博士后计划项目(2023ZB629)。

摘　　要：针对大深度载人舱球壳下潜过程难以推演等问题,本文提出一种数据驱动的下潜过程推演与异常诊断算法。首先,对大深度载人舱球壳结构及历史下潜数据进行分析。其次,将下潜深度作为输入,关键热点应力作为输出,利用长短时记忆神经网络(long short-term memory network,LSTM)构建下潜过程推演模型,并对推演结果进行分析。与DNN模型和BP模型进行对比,推演误差分别降低35.89%和68.30%。最后,基于LSTM模型,提出一种数据异常诊断算法,该算法可对传感器出现故障时的异常数据进行及时诊断与修正。Aiming at the difficulty in modelling the diving process of spherical shells,a data-driven algo⁃rithm for diving process modelling and anomaly detection of deep-sea pressurized spherical shells was pro⁃posed in this paper.Firstly,the spherical shell structures and historical diving data of manned capsules were analyzed.Then,the diving process modelling algorithm was established based on the long short-term memory network(LSTM),taking the diving depth as the input and the key hot spot strain as the output.The deduction results were analyzed and compared with the DNN model and BP model,the derivation error was reduced by 35.89%and 63.80%,respectively.Finally,based on the LSTM model,a data anomaly detection algorithm was proposed.The proposed algorithm can diagnose and correct abnormal data when a sensor fails.

关 键 词：数据驱动 载人舱球壳 长短时记忆神经网络 下潜过程推演 异常诊断 

分 类 号：P751[交通运输工程—港口、海岸及近海工程]