基于BP神经网络的桥梁移动荷载识别精度

为了识别作用于桥梁结构上的移动荷载,基于反向传播神经网络方法,开展了输入参数对荷载识别精度影响的分析.首先利用ANSYS模拟移动集中力通过简支T梁桥,得到了主梁跨中位移、速度和加速度时程曲线;其次基于MATLAB建立反向传播神经网络结构,分别将桥梁结构的位移、速度和加速度动态响应数据作为反向传播神经网络的输入参数,移动荷载大小作为输出参数,研究不同输入参数对荷载识别精度的影响;然后分别选取位移和速度、位移和加速度、速度和加速度以及三者组合的工况进行多参数输入的优化设计;最后,以某4跨预应力混凝土连续T梁桥工程为背景,以重车下的竖向加速度实测数据验证了该反向传播神经网络用于识别实桥上简单移动荷载的可行性.结果 表明:利用反向传播神经网络进行移动荷载大小识别时,单输入参数的识别精度由高到低依次为加速度、速度、位移,建议在实际工程中采用较易获取的加速度数据作为输入参数进行荷载识别;多参数组合输入可以提高移动荷载的识别精度,其中速度和加速度组合可以实现较优的识别效果;实测数据证明了该反向传播神经网络用于简单的实桥荷载识别是可行的.相关研究结果可为桥梁载荷识别及桥梁结构的性能评价提供参考.

Precision Analysis for Moving Load Identification of Bridge Structure Based on BP Neural Network

In order to identify the moving load acting on the bridge structure, the influence of input parameters on the accuracy of load identification was analyzed based on BP neural network method. Firstly, ANSYS was used to simulate the moving concentrated force passing the simply supported T-beam bridge, and the time history curves of displacement, velocity and acceleration in the middle of the span were obtained. Secondly, the BP neural network was established based on MATLAB, and the dynamic response data of displacement, velocity and acceleration of the structure were taken as the input parameters of BP neural network, and the moving load weight was taken as the output parameter to study the effect of input parameters on the accuracy of load identification. Then, the optimal design of multiple input parameters was carried out under the conditions of displacement and velocity, displacement and acceleration, velocity and acceleration and their combination. Finally, taking a four-span prestressed concrete continuous T-beam bridge as an example, the feasibility of the BP neural network to identify the simple moving load on the real bridge was verified by using the measured data of vertical acceleration under heavy vehicles. The results show that input parameter with the highest recognition accuracy is acceleration, followed by velocity and finally displacement when BP neural network is used to identify the moving load. It is suggested that the acceleration data which is easy to obtain should be used as the input parameter for load identification in real engineering. The multi-parameter combination input can improve the identification accuracy of moving load, and the combination of velocity and acceleration can achieve better identification result. The measured data show that the BP neural network is feasible for simple real bridge load identification. The results of this paper can provide reference for load identification and performance evaluation of bridge structure.