基于堆栈降噪自动编码器的桥梁损伤识别方法

Bridge Damage Identif ication Method Based on Stacked Denoising Autoencoders

基于深度学习理论,针对现有桥梁损伤模式识别法的不足,利用多个降噪自动编码器进行损伤特征的提取与组合,应用Softmax方法判断损伤模式,提出了基于堆栈降噪自动编码器的桥梁损伤识别方法。为了验证所提方法的准确性,以连续梁桥为例,使用所提方法及现有BP神经网络法进行损伤位置识别,对比了2种方法的识别精度和抗噪性能。研究结果表明:所提方法能准确识别损伤位置,相对于现有BP神经网络法具有更强的损伤识别能力、更高的识别精度及较强的抗噪能力。

In view of the shortcomings of existing pattern identification methods for bridge damage,a bridge damage identification method on the basis of stacked denoising autoencoder was proposed.Based on the theory of deep learning,multiple denoising autoencoders were applied to extract and combine the damage characteristics, then the damage mode was judged by Softmax method.Taking a continuous girder bridge as an example,damage location was identified by the proposed method and the existing method based on BP neural network in order to verify the effectiveness of the proposed method.Moreover, the identification accuracy and the anti-noise performance of these two methods were compared.The results indicate that the proposed method has stronger ability of damage identification and higher identification accuracy than the existing method based on BP neural network,and the anti-noise performance of the proposed method is good.