Abstract: In order to obtain the contemporary state for the cable replacement project of one certain existing single pylon cable-stayed bridge with double cable plane，a new combined method for finite element model updating is proposed． In the light of the parameterized model updating theory，the relative sensitivities of the calculated design parameters are analyzed to determine the will be modified parameters． Substructure method is introduced in the model updating process for meet the requirement of the parameter’s discreteness，and the calculated parameters in each substructure is regard as invariables． The radial basis function neural network ( ＲBF) is adopted as the optimization algorithm of model updating． Combination the substructure method and ＲBF，the intrinsic“inverse problem”of finite element model updating is transformed as the“forward problem”． The substructure partition，ＲBF neural network construction and its input and output parameters determination are discussed as well． A certain existing single pylon cable-stayed bridge is taken as the case study to verify the proposed combined model updating algorithm． The result shows that the discrepancy between the calculated value and measured valued decrease dramatically before and after the finite element model updating．

Key words: finite element model updating, radial basis function neural network, single pylon cable-stayed bridge, substruc-ture, relative sensitivity

摘要： 为获得某单塔双索面斜拉桥换索过程中的工作状态，建立了一种联合子结构与径向基神经网络的有限元模型修正新方法。根据模型参数修正理论，通过分析设计参数的相对灵敏度确定需要修正的参数; 为满足参数离散性要求，在模型修正过程中引入了子结构方法，并认为每一子结构中的设计参数是不变的。采用径向基( ＲBF) 神经网络作为模型修正优化算法。将子结构与 ＲBF 神经网络相结合，从而将有限元模型修正的反问题转化为正问题; 同时，对子结构的划分、ＲBF 神经网络构建以及输入输出参数的确定进行了讨论。以某单塔斜拉桥为例，验证了所提的联合模型修正方法。结果表明: 计算值与测量值之间的误差，在有限元模型修正前后有很大改善。

关键词: 有限元模型修正, 径向基神经网络, 单塔斜拉桥, 子结构, 相对灵敏度