基于遗传算法的BP神经网络对海底管道受撞击损伤预测
|
摘要
人类频繁的海洋活动中难免发生重物落水事故,对海底管道造成撞击损伤,引起环境污染及经济损失。为保证管道在运行期间的安全性,有必要准确快速的对管道损伤进行预测以便为实际工程提供参考。BP神经网络常作为损伤预测的一种数学模型,但本身易陷入局部极小且预测精度较低。针对上述问题,本文提出了基于遗传算法的BP神经网络(GA-BP神经网络)损伤预测模型。利用有限元计算数据构成样本空间,对管道损伤进行预测,并将结果与BP神经网络、有限元计算的结果进行对比。分析表明:与BP神经网络相比,GA-BP神经网络的预测结果与有限元计算的结果较为接近,预测精度较高,其平均误差为1.27%,满足工程精度要求的同时又节省了计算时间。
With the increase of human activities on the ocean, it is inevitable that weights drop into the sea, which may lead to damages of the pipelines once impacted. The failure of pipelines may cause great financial losses and serious environment pollution. In order to ensure the safety of pipelines during operation period, it is necessary to predict the damage fast and accurately, which can provide reference for engineering practice. Back-propagation neural network(hereinafter BP neural network) is a common mathematic model to predict the pipeline damage. However, it is prone to plunge into local minimum which leads to errors. To solve the problem, this paper proposes a modified BP neural network model to predict the pipeline damage based on genetic algorithm(hereinafter GA-BP neural network). Sample space for the construction of the proposed model is consisted of the data by finite element method(hereinafter FEM). Subsequently, damage predictions are made and the results are compared with that by BP neural network and by FEM. The analysis indicates that, compared with the BP neural network, the GA-BP neural network has higher prediction accuracy and are more close to the FEM results. Besides, the average relative error is around 1.27%. Therefore, the engineering accuracy requirements can be satisfied and the considerable calculation time is saved.
With the increase of human activities on the ocean, it is inevitable that weights drop into the sea, which may lead to damages of the pipelines once impacted. The failure of pipelines may cause great financial losses and serious environment pollution. In order to ensure the safety of pipelines during operation period, it is necessary to predict the damage fast and accurately, which can provide reference for engineering practice. Back-propagation neural network(hereinafter BP neural network) is a common mathematic model to predict the pipeline damage. However, it is prone to plunge into local minimum which leads to errors. To solve the problem, this paper proposes a modified BP neural network model to predict the pipeline damage based on genetic algorithm(hereinafter GA-BP neural network). Sample space for the construction of the proposed model is consisted of the data by finite element method(hereinafter FEM). Subsequently, damage predictions are made and the results are compared with that by BP neural network and by FEM. The analysis indicates that, compared with the BP neural network, the GA-BP neural network has higher prediction accuracy and are more close to the FEM results. Besides, the average relative error is around 1.27%. Therefore, the engineering accuracy requirements can be satisfied and the considerable calculation time is saved.
引文
[1] Risk Assessment of Pipelines Protection(DNV-RP-F107)[S],Oslo:Det Norske Veritas,2010.
[2] Submarine pipeline system(DNV-OS-F101)[S],Oslo:Det Norske Veritas,2013.
[3] Yong Bai,Preben Terndrup Pedesren.Elastic-plastic behaviour of offshore steel structures under impact loads[J].England:International Journal of Impact Engineering,1993,13(1):99-115.
[4] 粟京.DNV96版《海底管道系统规范》对冲击防护的新规定[J].西安,石油工业技术监督,15(2):15-17.
[5] 黄小光,孙峰.基于ANSYS/LS-DYNA的海底管道受抛锚撞击动力学仿真[J].上海,中国海洋平台,2012,27(5):41-44.
[6] 白俊磊.海底管道坠物碰撞损伤数值模拟分析研究[D].大连:大连理工大学,2013.
[7] 杨慧娟,闫涛,修宗祥,等.海底管道受坠物撞击时的弹塑性有限元分析[J].北京:工程力学,2011,28(6):189-194.
[8] 娄敏,明海芹.基于LS-DYNA海底悬空管道受坠物碰撞动力响应分析[J].天津,海洋通报,2015,34(1) :113-120.
[9] 张向东,董胜,张磊,等.防波堤的人工神经网络Monte Carlo法可靠性分析[J].青岛:中国海洋大学学报,2012,42(4):83-86.
[10] Hornik K,Stinchcomebe M,White H.Multilayer feedforward networks are universal approximators[J].England:Neural networks,1989,2(5):359-366.
[11] Hornik K,Stinchcomebe M,White H.Universal approximation of an unknown mapping and its derivatives using multilayer feedforward network[J].England:Neural networks,1990,3(5):551-560.
[12] MATLAB中文论坛.MATLAB神经网络30个案例分析[M].北京:北京航空航天大学出版社,2010.
[13] 石亦平,周玉蓉.ABAQUS有限元实例详解[M].北京:机械工业出版社,2006.
[14] 王自力,蒋志勇,顾永宁.船舶碰撞数值仿真的附加质量模型[J].绵阳:爆炸与冲击,2002,22(4):321-326.
[15] 曹金凤,王旭春,孔亮.Python语言在Abaqus中的应用[M].北京:机械工业出版社,2011.
[16] 周毅,徐伯龄.神经网络中的正交设计法研究[J] .南京:南京大学学报(自然学科版),2001,37(1):72-78.
[2] Submarine pipeline system(DNV-OS-F101)[S],Oslo:Det Norske Veritas,2013.
[3] Yong Bai,Preben Terndrup Pedesren.Elastic-plastic behaviour of offshore steel structures under impact loads[J].England:International Journal of Impact Engineering,1993,13(1):99-115.
[4] 粟京.DNV96版《海底管道系统规范》对冲击防护的新规定[J].西安,石油工业技术监督,15(2):15-17.
[5] 黄小光,孙峰.基于ANSYS/LS-DYNA的海底管道受抛锚撞击动力学仿真[J].上海,中国海洋平台,2012,27(5):41-44.
[6] 白俊磊.海底管道坠物碰撞损伤数值模拟分析研究[D].大连:大连理工大学,2013.
[7] 杨慧娟,闫涛,修宗祥,等.海底管道受坠物撞击时的弹塑性有限元分析[J].北京:工程力学,2011,28(6):189-194.
[8] 娄敏,明海芹.基于LS-DYNA海底悬空管道受坠物碰撞动力响应分析[J].天津,海洋通报,2015,34(1) :113-120.
[9] 张向东,董胜,张磊,等.防波堤的人工神经网络Monte Carlo法可靠性分析[J].青岛:中国海洋大学学报,2012,42(4):83-86.
[10] Hornik K,Stinchcomebe M,White H.Multilayer feedforward networks are universal approximators[J].England:Neural networks,1989,2(5):359-366.
[11] Hornik K,Stinchcomebe M,White H.Universal approximation of an unknown mapping and its derivatives using multilayer feedforward network[J].England:Neural networks,1990,3(5):551-560.
[12] MATLAB中文论坛.MATLAB神经网络30个案例分析[M].北京:北京航空航天大学出版社,2010.
[13] 石亦平,周玉蓉.ABAQUS有限元实例详解[M].北京:机械工业出版社,2006.
[14] 王自力,蒋志勇,顾永宁.船舶碰撞数值仿真的附加质量模型[J].绵阳:爆炸与冲击,2002,22(4):321-326.
[15] 曹金凤,王旭春,孔亮.Python语言在Abaqus中的应用[M].北京:机械工业出版社,2011.
[16] 周毅,徐伯龄.神经网络中的正交设计法研究[J] .南京:南京大学学报(自然学科版),2001,37(1):72-78.