基于LSTM网络的箱梁桥应变与挠度的非线性映射模型研究

Research on Nonlinear Mapping Model of Strain and Deflection of Box-Girder Bridge Based on LSTM Networks

以烈士河大桥(12联混凝土简支箱梁桥)为例,基于长短期记忆(LSTM)网络建立了应变和挠度监测数据间的非线性映射模型。首先,运用小波包分解与重构方法分离实测应变数据中的温度应变,得到车载应变数据。然后,运用LSTM网络分别构建应变和挠度的单对单和多对多映射模型。结果表明,单对单和多对多映射模型的相关性和误差均满足控制标准;多对多映射模型的回归预测偏差较单对单映射模型更小,性能优于单对单映射模型。

Taking Martyr River Bridge(a 12-box simply supported concrete box-girder bridge)as an example,a nonlinear mapping model of strain and deflection is established based on long short-term memory(LSTM)networks.First,the method of wavelet packet decomposition and reconstruction is used to separate the temperature-induced strain component to obtain the vehicle-induced strain.Then,the single-to-single and many-to-many mapping models of strain and deflection are constructed by using LSTM networks.The results show that the correlation and error of both the single-to-single and many-to-many mapping models meet the requirement of the control standards,the regression prediction error of the many-to-many mapping model is smaller than that of the single-to-single mapping model,and the performance of the many-to-many mapping model is better than that of the single-to-single mapping model.