摘要
针对有限元模型、风洞试验等难以实时预测风振响应的问题,提出基于稀疏高斯过程回归的强/台风作用下大跨度桥梁风振响应概率预测方法。该方法从数据驱动的角度出发,采用稀疏近似方法降低常规高斯过程模型存储空间,将风特性参数与风振响应的历史监测数据同时作为输入变量,并根据联合假设检验比较各变量的重要性程度以确定最终输入特征,进而实现风振响应的动态预测。采用苏通大桥2008年至2012年的七次台风数据对该方法的预测精度与效率进行验证。结果表明:稀疏高斯过程回归相对于常规高斯过程可有效减少模型训练时间;除风特性参数外,在模型的输入变量中考虑风振响应历史数据可进一步提高预测精度;相较于随机森林算法和多元线性回归,稀疏高斯过程回归表现出更好的预测性能。
A probabilistic method for predicting the typhoon-induced responses is proposed considering that the finite element model and wind tunnel test may fail to predict the wind effect in real-time.From the data-driven perspective, the approximate sparse strategies are employed to decrease the computational storage of the conventional Gaussian process model.With the the wind characteristic parameters and the historical data of wind effect taken as the input variables, the F-test is adopted to compare the importance of various inputs so that the final input features can be determined for the dynamic prediction on the typhoon-induced responses.The monitoring data of seven typhoons of the Sutong Cable-Stayed Bridge from 2008 to 2012 are utilized to validate the computational efficiency and accuracy of the proposed method.The results indicate that the sparse Gaussian process regression can efficiently reduce the time of model training.Apart from the wind characteristic parameters, considering the historical data of wind effect in the input of model can further improve the prediction accuracy.The proposed method exhibits the better accuracy compared to the random forest algorithm and the multivariable linear regression.
作者
张一鸣
王浩
茅建校
Zhang Yiming;Wang Hao;Mao Jianxiao(Key Laboratory of Concrete and Prestressed Concrete Structure of Ministry of Education,Southeast University,Nanjing 211189,China)
出处
《土木工程学报》
EI
CSCD
北大核心
2022年第10期72-79,共8页
China Civil Engineering Journal
基金
国家自然科学基金(51978155、52108274)
江苏省重点研发计划(BE2018120)。
关键词
风振响应
概率预测
强/台风
大跨度桥梁
稀疏高斯过程回归
wind effects
probabilistic prediction
strong wind/typhoon
long-span bridges
sparse Gaussian process regression
