基于神经网络的桥梁异常状态识别方法  

A Bridge Abnormal State Recognition Method Based on Neural Network

作　　者：柏平 马志华 姜世英 华浩 李文华 周晋宇[3,4] 王鹏军[3,4] 

机构地区：[1]南通市公路管理处,江苏南通226001 [2]无锡市政设施管理处,江苏无锡214002 [3]无锡源清慧虹信息科技公司,无锡214000 [4]清华大学,北京100084

出　　处：《上海公路》2015年第3期36-39,50,共5页Shanghai Highways

基　　金：江苏公路交通科研计划项目(2014Y02)

摘　　要：针对桥梁结构监测采集到的桥梁异常状态下长期积累演变的惊人数据量,提出了基于主成分分析与人工神经网络相结合的桥梁结构异常状态识别方法。布设多种类型传感器监测获取高维数据,采用主成分分析法对原始高维特征数据进行预处理,将结构异常特征变量的主成分作为人工神经网络的输入特征。该方法有效的降低了神经网络的结构复杂度,同时提高人工神经网络的训练速度,也保证了人工神经网络具有良好的收敛性和稳定性。应用于江苏南通如泰运河大桥和江苏无锡开源大桥的实际监测数据的结果表明,基于主成分分析的人工神经网络方法用于桥梁结构异常状态识别,与传统的神经网络以及其他模式识别算法相比,有更好的识别精度。By integrating artificial neural network(ANN) and principal component analysis(PCA),a compound method using massive monitoring data for abnormal state recognition of bridge strcuture is proposed.Multiple type sensors are deployed to monitor for high dimensional data.The PCA method is applied to acquire low-dimensional features from high-dimensional raw sensor inputs,helping reduce the scale and complexity of the ANN and improve network training speed to ensure that the ANN method has good convergence and stability.The applications of the method in Jiangsu Nantong Rutai Bridge and Jiangsu Wuxi Kaiyuan Bridge show the result that the compound method has a better recognition accuracy compared with ordinary backward-propagated neural networks and other pattern recognition algorithms.

关 键 词：人工神经网络 主成分分析 结构监测 模式识别 

分 类 号：U446[建筑科学—桥梁与隧道工程]