基于机器学习的三维隔震体系地震峰值响应预测与重要度分析
Machine Learning-based Prediction and Importance Analysis of Peak Seismic Responses for Three-dimensional Isolation Systems
摘要
为提高三维隔震结构地震响应的分析效率,以单质点动力体系作为研究对象,利用有限元模拟与机器学习,开展峰值响应预测和参数敏感性研究.基于 OpenSee 建立三维隔震单质点体系的有限元模型.选取FEMA P-695 推荐的32 条地震动记录作为输入,利用Python 编程实现了参数自动组合、有限元模型批量生成、地震动自动读取以及多工况非线性时程分析,构建机器学习训练样本数据库.采用随机森林、XGBoost、LightGBM 和 CatBoost 四种典型机器学习模型,对结构峰值加速度和峰值位移进行回归预测.以决定系数、均方根误差和平均绝对误差评价了模型预测性能,并结合置换重要度分析了输入参数的敏感性.预测结果表明:XGBoost 模型在四种机器学习模型中的回归性能最优,可在保证预测精度的前提下有效提升三维隔震体系地震峰值响应的分析效率.参数敏感性分析进一步揭示,FPB 与 DPS-DB 串联配置下的水平-竖向耦合效应导致水平隔震参数对竖向响应具有显著的影响.
The nonlinear time-history analysis of structures under seismic excitation typically relies on refined finite ele-ment models,which needs substantial computational effort and time.To improve the efficiency of seismic response analy-sis for three-dimensional(3D)isolation systems,this work focuses on a lumped-mass model and integrates finite element simulations with machine learning methods to predict peak responses and investigate parameter sensitivity.The construc-tion and dynamic modeling of the 3D isolation device are described,and a finite element model of the 3D isolated lumped-mass system is established in OpenSees.A total of 32 ground motion records recommended by FEMA P-695 are selected as dynamic excitation.A Python-based automated workflow is developed to perform parameter permutation,batch model generation,ground motion loading,and nonlinear time-history analysis,thereby constructing the training da-tabase.Four machine learning models,such as random forest,XGBoost,LightGBM,and CatBoost,are employed to pre-dict peak structural accelerations and displacements.Model performance is assessed using the coefficient of determina-tion,root mean square error,and mean absolute error,while permutation importance is applied to evaluate input parame-ter sensitivity.The prediction results indicate that XGBoost exhibits superior regression performance,enabling an effi-cient and reliable prediction of peak seismic responses for 3D isolation systems without compromising accuracy.The pa-rameter sensitivity analysis further reveals that the horizontal-vertical coupling effect inherent in the series configuration of FPB and DPSDB results in a significant influence of horizontal isolation parameters on the vertical response.
年,卷(期):
2026, 24(4)
分类号:
TU352.1
机标分类号:
TU352.12
TP391.4
P315.9
在线出版日期:
2026年8月25日
基金项目:
2022MFSTL12:北京建筑大学大型多功能振动台阵实验室开放研究专项基金8182016:北京市自然科学基金资助项目,