基于RBFNN-ISSA的特大跨径悬索桥有限元模型修正  

Finite element model correction of super long-span suspension bridge based on RBFNN-ISSA

作　　者：王祺顺 何维 吴欣 郭伟奇 雷顺成 WANG Qishun;HE Wei;WU Xin;GUO Weiqi;LEI Shuncheng(Hunan Key Laboratory of Transportation Construction Engineering,Hunan Transportation Science Research Institute Co.,Ltd.,Changsha 410015,China;Key Laboratory of Bridge Engineering Safety Control,Provincial and Ministry of Education,School of Civil Engineering,Changsha University of Technology,Hunan 410015,China;Hunan Key Laboratory of Wind Engineering and Bridge Engineering,College of Civil Engineering,Hunan University,Changsha 410082,China)

机构地区：[1]湖南省交通科学研究院有限公司,交通建设工程湖南省重点实验室,长沙410015 [2]长沙理工大学土木工程学院桥梁工程安全控制省部共建教育部重点实验室,长沙410015 [3]湖南大学土木工程学院风工程与桥梁工程湖南省重点实验室,长沙410082

出　　处：《振动与冲击》2024年第7期155-167,共13页Journal of Vibration and Shock

基　　金：国家自然科学基金青年项目(52108139);湖南省科技厅重点领域研究计划(2019SK2171);交通运输行业重点科技项目(2021-ZD7-098);湖南省交通科技项目(202045);湖南省交通科技项目(202206)。

摘　　要：针对大跨径悬索桥一类复杂结构的有限元模型修正问题,提出了一种基于径向基神经网络(radial basis function neural network,RBFNN)子结构代理模型与改进麻雀搜索算法(improved sparrow search algorithm,ISSA)的有限元模型修正方法。首先,基于桥梁图纸数据采用通用有限元软件建立一座大跨悬索桥的初始有限元模型,并根据拉丁超立方抽样原则生成子结构材料参数-结构响应的训练样本,通过RBF神经网络和子结构模拟方法对初始有限元模型进行解构重组和样本学习,拟合关于材料参数-结构响应的代理模型。其次,建立考虑主梁挠度和模态频率误差最小的有限元模型参数修正数学优化模型,采用Tent混沌映射及黄金正弦策略改进标准麻雀搜索算法,引入柯西分布函数和贪心保留策略对每一代麻雀种群进行扰动,以用于求解联合静、动力特征的有限元模型修正数学优化问题。最后,以杭瑞高速洞庭湖大桥为工程背景,进行了悬索桥荷载试验,利用实测桥梁响应数据验证了该方法的可行性。研究结果表明:基于RBF神经网络与子结构法的模型修正方法,可以建立拟合精度较高的悬索桥结构代理模型;基于子结构RBF神经网络与改进麻雀搜索算法修正后的有限元模型相较于整体RBF神经网络、支持向量机和Kriging模型,大幅提升了对于实际结构的模拟精度,与实测数据相比,修正前后有限元模型在两级静力加载工况下13个有效测点挠度的平均相对误差降低了25%以上,前8阶模态频率的平均相对误差由-6.83%降至-2.38%,MAC值结果表明修正后模型能够准确地反映出大桥的实际振动状态,有效改善了初始有限元模型计算失真的情况;此外,基于混合策略改进后的麻雀搜索算法对于有限元模型修正参数的寻优具有更佳的收敛效率和稳定性。Here,aiming at the problem of finite element model correction for a class of complex structures of long-span suspension bridges,a finite element model correction method based on radial basis function neural network(RBFNN)substructure surrogate model and improved sparrow search algorithm(ISSA)was proposed.Firstly,based on bridge drawing data,an initial finite element model of a large-span suspension bridge was established using general finite element software.According to Latin hypercube sampling principle,training samples for substructure material parameters-structural response were generated.The initial finite element model was deconstructed and reconstructed for sample learning with RBFNN and substructure simulation method to fit a surrogate model for material parameters-structural response.Secondly,a finite element model parametric correction mathematical optimization model considering minimum errors for girder deflection and modal frequencies was established.Tent chaotic mapping and golden sine strategy were used to improve the standard sparrow search algorithm,Cauchy distribution function and greedy preservation strategy were introduced to perturb each generation of sparrow population,and solve the finite element model correction mathematical optimization problem for joint static-dynamic characteristics.Finally,taking Dongting Lake bridge on Hang-Rui Expressway as the engineering background,suspension bridge load tests were conducted,and actually measured bridge response data were used to verify the feasibility of the proposed method.The study results showed that the proposed model correction method based on RBFNN and substructure method can establish a suspension bridge structure surrogate model with higher fitting accuracy;compared with the overall RBFNN,support vector machine and Kriging model,the finite element model modified based on substructure RBFNN-ISSA can significantly improve the simulation accuracy of actual structures;compared with actually measured data,the average relative error of 13 effecti

关 键 词：桥梁工程 有限元模型修正 改进麻雀搜索算法(ISSA) 悬索桥 径向基神经网络(RBFNN) 柯西变异策略 

分 类 号：U448.25[建筑科学—桥梁与隧道工程]