摘要
在建筑抗震工程领域,引入机器学习算法能高效、准确地解决复杂工程的抗震问题,故开发了能够预测大跨钢结构在地震作用下的结构动力响应快速计算代理模型。以某大跨钢结构厂房为研究对象,通过OpenSees软件开展100条地震动作用下的结构非线性时程分析,记录结构动力响应,据此构造用于训练代理模型的虚拟数据池。采用BP神经网络和集成学习(XGBoost)算法,分别对大跨钢结构的地震响应数据集进行深度学习和统计分析,描述网络输入特征值与输出值的非线性行为。结果表明:用少量模型计算后的样本数据集进行机器学习训练的预测模型可以快速预测大跨钢结构的响应,BP神经网络模型的预测精度高于XGBoost算法模型,采用BP神经网络模型时应警惕网络隐藏层数目过多引发的过拟合问题。结构地震响应代理模型适用于总跨度50~70m、平面长宽比≤5∶1的同类型单层钢结构厂房,可显著减少结构非线性动力时程的计算负担,同时有助于大跨钢结构的地震韧性快速评估。
In the field of building seismic engineering,the introduction of machine learning algorithms can efficiently and accurately solve the seismic problems of complex projects.Therefore,a fast calculation surrogate model that can predict the structural dynamic response of large-span steel structures under earthquake was developed.Taking a large-span steel structure workshop as the research object,the nonlinear time-history analysis of the structure under 100 ground motions was carried out by OpenSees software,and the dynamic response of the structure was recorded.Based on this,a virtual data pool for training the surrogate model was constructed.The BP neural network and extreme gradient boosting(XGBoost)algorithm were used to perform deep learning and statistical analysis on the seismic response data set of large-span steel structures,respectively,to describe the nonlinear behavior of network input eigenvalues and output values.The results show that the prediction model trained by machine learning with a small number of sample data sets calculated by the model can quickly predict the response of large-span steel structures.The prediction accuracy of the BP neural network model is higher than that of the XGBoost algorithm model.When using the BP neural network model,it is necessary to be alert to the over-fitting problem caused by the excessive number of hidden layers in the network.The seismic response surrogate model is suitable for the same type of single-layer steel structure workshops with a total span of 50~70m and a plane aspect ratio of≤5∶1,which can significantly reduce the computational burden of structural nonlinear dynamic time history.At the same time,it is helpful for the rapid evaluation of seismic resilience of large-span steel structures.
作者
王伟伟
WANG Weiwei(China Railway 14th Bureau Group Construction Engineering Co.,Ltd.,Jinan 250200,China)
出处
《建筑结构》
北大核心
2025年第24期103-109,共7页
Building Structure
基金
山东省重点研发计划(重大科技创新工程)-绿色智能建造和建筑工业化关键技术、成套装备及应用(2021CXGC011204)。
关键词
大跨钢结构
BP神经网络
集成学习算法
快速计算
代理模型
地震韧性
large-span steel structure
BP neural network
extreme gradient boosting algorithm
rapid calculation
surrogate model
seismic resilience
