摘要
半潜式平台承受着风、浪、流等复杂环境荷载的耦合作用,在工作海况下平台的浮体运动多为波频小幅运动。在极端海况下平台产生大幅运动对结构的安全带来威胁。本文基于深度学习理论,开展了半潜式平台运动响应预测及分布规律的研究。首先,按照10 min为时间间隔对环境监测信息进行划分,对风速、波浪压力等环境监测信息的分布规律进行研究并选取合适的分布拟合参数,结合分形学理论及统计分析的方法,提出了实测风速、波浪压力等数据的特征参数,并结合浪高、周期、流速、流向等实测数据,建立了具有降维特征的环境信息输入参数;其次,基于实测响应数据,以横摇为例,以10 min为时间间隔对其监测信息进行划分并对其分布规律进行研究,并选取合适的响应分布拟合参数作为响应的特征参数;接着,利用北斗远程传输系统传输的监测数据,基于深度置信神经网络(Deep Belief Network,DBN)建立极端海况下实测环境荷载与实测响应的关系模型,并与基于BP、Elman神经网络的关系模型预测结果进行对比,可以看出,基于DBN神经网络的关系模型预测误差仅为5.07%,结果较为准确;最后,基于DBN神经网络建立了荷载特征参数与响应分布拟合特征参数的关系模型,并与基于DNN、BP神经网络的关系模型预测结果进行对比。研究发现,基于DBN神经网络的预测模型结果更为准确,更接近于真实响应的分布规律,可以对工作海况下平台安全作业提供一定的指导。
A semi-submersible platform is subjected to the coupling interactions of complex load environments such as wind,wave and current.In a majority of sea conditions,the movements of a floating platform are usually small wave-induced motions.In extreme conditions,the large movements generated by the platform pose a great threat to the safety of the platform structure.Based on deep learning theory,the research on the motion response prediction and distribution of semi-submersible platforms is presented in this paper.Firstly,the environment monitoring information is divided with a time interval of 10 minutes and the distribution fitting parameters of load monitoring information such as wind speed are studied.By using fractal theory and statistical analysis methods,the characteristic parameters of the measured load data are extracted.Combined with the measured data of wave height,period,current velocity and current direction,the input parameters of environment information with reduced dimensions are established.Secondly,based on the measured response data,with the rolling taken as an example,the monitoring information is divided with a time interval of 10 minutes.The distribution is studied and the appropriate response distribution fitting parameters are selected as the response characteristic parameters.Then,the relationship model between the measured environmental load and the measured response under survival condition is established based on the monitoring data transmitted by Beidou remote transmission system and the deep belief network(DBN),and the prediction results are compared with those based on BP and Elman neural network.It can be seen that the prediction error of the relationship model based on the DBN network is only 5.07%,which is more accurate.Finally,the relationship model between load characteristic parameters and response distribution fitting characteristic parameters is established based on DBN neural network,and the prediction results are compared with those based on DNN and BP neural network.It can be seen that prediction results based on the DBN network are more accurate and closer to the real corresponding distribution law,which can provide certain guidance for the safe operation of the platform under various sea conditions.
作者
陈海
李志刚
冯加果
CHEN Hai;LI Zhi-gang;FENG Jia-guo(Economics and Management School,Beijing University of Posts and Telecommunications,Beijing 100876,China;CNOOC China Limited,Beijing 100010,China)
出处
《船舶力学》
EI
CSCD
北大核心
2021年第5期586-597,共12页
Journal of Ship Mechanics
基金
工信部创新专项、国家科技重大专项(2016ZX05028-002)
国家重点研发计划项目(2016YFC0303600)。
关键词
海洋环境荷载特征参数
响应分布拟合参数
深度置信网络
ocean environment load characteristic parameters
response distribution fitting parameters
deep belief network
