基于改进的BP神经网络的钢桁梁桥损伤识别

文章对某钢桁梁黄河大桥进行了损伤数值模拟,提取其固有频率作为BP神经网络的输入参数来训练网络,对桥梁整体的损伤进行诊断,并根据实桥损伤诊断的结果提出了一种改进的BP神经网络方法,它能够解决传统BP算法的梯度下降速度,从而提高运算速度,通过自调节保证学习过程中每一时刻具有较大的Sigmoid函数值,避免了局部极小。

Damage identification of a steel truss bridge based on the improved BP neural network

Theoretically the locations and degree of structural damage of a bridge can be reflected by the changes in its natural frequencies before and after damage.In this paper,the damage degree of a steel truss bridge is simulated with the numerical method,and the natural frequencies are used as the network input parameters to train the back-propagation(BP) network so as to diagnose the bridge's global damage.An improved BP neural network is proposed to solve the problem of the traditional BP method and to enhance computational efficiency,and the local minimum is avoided by keeping the Sigmoid derivative relatively larger.