The seismic observation data of the dam reflect the response of the ground at the strong earthquake monitoring site of the dam under earthquake action,which is an important characteristic of the dynamic characteristics of the structure.At present,the application of dam seismic data analysis is limited to the acquisition of the acceleration amplification factor and the structural modal parameter identification.After identifying the modal parameters of the dam according to the strong motion data,the dynamic material parameters of the structure can be back-analyzed so as to provide an accurate numerical model for the anti-earthquake capacity analysis of the structure.For the dynamic material parameters back-analysis of large hydraulic structures,such as dams,it is necessary to take the computational efficiency of a method into account.In this work,we put forward the advantage of the machine learning algorithm based on the multi-output support vector machines (M-SVM) model instead of finite element calculation,in order to reduce the complexity of finite element calculation in the process of optimization,and improve the efficiency of the calculation.The simulated data of a numerical example and the strong-motion record of a realistic engineering problem are both analyzed using the proposed back-analysis method.It shows that the back-analysis method has good accuracy and high computation efficiency and can be applied to practical engineering problems.%大坝的强震观测资料真实反映了地震作用下坝体强震监测部位的反应情况,是认识结构动力特性的重要特征量.目前,大坝强震观测资料分析的应用研究一般仅局限于获取加速度放大系数和结构模态参数.根据强震观测识别出大坝模态参数后,可进行结构材料动参数的反演,以便为结构抗震分析提供尽可能精确的数值模型,对大坝等大型水工结构进行动参数反演,需要考虑计算效率的问题.为此,提出了应用机器学习的优势算法,即多输出支持向量机(M-SVM)模型,替代有限元计算,以减少优化搜索过程中有限元计算的次数,提高计算的效率.通过对数值仿真算例和某混凝土重力坝实际强震观测数据的分析,表明所提出的基于强震观测和M-SVM模型的大坝材料动参数反演方法具有精度高和计算效率高等优点,在实际工程中有良好的应用前景.
展开▼
