摘要
为提高桥梁地震响应预测效率,提出一种基于可解释机器学习的连续梁桥地震预测方法。该方法以公路网典型四跨连续梁桥结构特性和地震动参数为基础,首先构建桥梁设计特征参数和地震响应的数据集;然后运用决策树、支持向量机、随机森林及人工神经网络等机器学习模型对数据集进行训练和测试,采用决定系数R2、均方根误差RMSE、平均绝对误差MAE和均方误差MSE等指标评估模型的预测精度;最后基于可解释技术对最优模型中影响桥梁抗震性能的主要设计参数进行分析,并将最优模型与传统精确模拟进行对比。结果表明:人工神经网络模型对连续梁桥地震响应预测的适用性最高,其R2达到0.98,且RMSE、MAE和MSE均较小;地震动峰值加速度和桥墩直径是影响预测结果的主导因素,地震动峰值加速度、墩高、跨径与预测结果呈正相关,桥墩直径与其呈负相关;人工神经网络模型预测结果与传统精确模拟的误差在2.5%以下,且计算效率提高了500多倍。
To improve the efficiency of seismic response prediction for continuous girder bridges in highway networks,this paper presents a prediction method based on explainable machine learning.Using the structural characteristics of a typical four-span reinforced concrete continuous girder bridge in a highway network and seismic motion parameters as the foundation,a training dataset containing bridge design feature parameters and seismic responses is built first.Machine learning methods including decision trees,support vector machines,random forests,and artificial neural networks are then employed to establish nonlinear mapping relationships between bridge design features and seismic responses,with Bayesian optimization and cross-validation ensuring model performance.The prediction accuracy of different machine learning models is evaluated using metrics such as the coefficient of determination(R 2),root mean square error(RMSE),and maximum absolute error(MAE).Finally,interpretability analysis of the key design parameters affecting bridge seismic performance in the optimal model is conducted using score addition interpretation and partial dependence analysis techniques,and the best-trained machine learning model was validated against traditional precise simulation.Results show that the artificial neural network model demonstrates the highest applicability for predicting seismic responses of continuous girder bridges,achieving an R 2 of 0.98 and smaller RMSE,MAE and MSE.Peak ground acceleration(PGA)and pier diameter are identified as the dominant factors affecting prediction results,with PGA,pier height,and span length showing positive correlations with prediction results,while pier diameter exhibiting a negative correlation.The error between artificial neural network predictions and traditional precise simulation is below 2.5%,with computational efficiency improved by more than 500 times.
作者
李悦
晏勇
张常勇
李冲
李淑明
LI Yue;YAN Yong;ZHANG Changyong;LI Chong;LI Shuming(School of Civil Engineering,North China University of Technology,Beijing 100144,China;Shandong Provincial Communications Planning and Design Institute Group Co.,Ltd.,Jinan 250031,China;CCCC Highway Bridges National Engineering Research Centre Co.,Ltd.,Beijing 100120,China;Zhongyu Tiexin Transport Technology Co.,Ltd.,Hengshui 053020,China)
出处
《世界桥梁》
北大核心
2025年第4期69-77,共9页
World Bridges
基金
国家自然科学基金项目(51408009)
北京市属高校基本科研业务费项目(110052971921/062)。
关键词
公路桥
连续梁桥
地震响应
设计特征参数
机器学习
可解释技术
人工神经网络模型
预测精度
highway network
continuous girder bridge
seismic response
structural characteristic parameter
machine learning
interpretable machine learning
artificial neural network
prediction precision
