结合小波低频子带的SVM桥梁挠度修正方法  

SVM Bridge Deflection Modification Method Incorporating Wavelet Low-Frequency Band

作　　者：胡顺仁[1,2] 熊文超[1] 张建科[1] 甘亚晨 李瑞平[1] 

机构地区：[1]重庆理工大学电子信息与自动化学院,重庆400054 [2]重庆大学光电技术及系统教育部重点实验室,重庆400044

出　　处：《世界桥梁》2014年第1期71-76,共6页World Bridges

基　　金：国家科技部863项目(2006AA04Z433);国家科技部重大攻关资助项目(2002BA105C);重庆市科技攻关重大专项(7289);重庆市自然基金资助项目(2005BB6081)

摘　　要：使用支持向量机进行桥梁挠度修正时，若样本数据量较大，运算速度会较慢，为解决该问题，提出一种结合小波低频子带的挠度数据预处理方法，该方法通过选择合适的小波参数，将挠度传感器数据转换到小波低频子带进行预处理，再作为支持向量机的样本数据进行挠度修正，然后通过小波重构得到挠度传感器的理论值，将其代入公式即可得到修正后的挠度值。试验分别选取某桥300个样本数据进行学习和训练，经预处理后数据量仅为43个，运算时间从原来的33S降低到0．1～0．2S，表明修正计算的运算时间大幅降低；同时挠度均方误差由原来的0．3349降低到0．280，表明修正精确度略有提高，证明该方法具有很好的实用性。When the Support Vector Machine （SVM） is used to modify the bridge deflection, if the number of the sample data is enormous, the calculation speed would be slow. In order to solve the problem, a kind of deflection data processing method incorporating the wavelet low-frequency band was put forward. By selecting proper wavelet parameters, this method can transfer the data in the deflection sensors to the wavelet low-frequency band for preprocessing. Then the data can be used as the sample data of SVM for deflection modification, and the next step is to gain the theoretical values of the deflection sensors by wavelet reeomposition, and substitute the theo- retical values to the formulae to get the deflection values after the modification. A number of 300 sample data of a bridge were selected for the study and training in the testing. After the prepro- cessing, the number of data was cut down to 43, the calculation time declined from the initial 33 s to 0.1~0.2 s, which demonstrated that the calculation time for the modification calculation was significantly decreased. Meanwhile, the mean squared errors of deflection decreased from the ini- tial 0. 334 9 to 0. 280, which indicated that the accuracy of the modification had been improved a little and the method has convincing practicality as well.

关 键 词：桥梁工程 数据预处理 小波低频子带 支持向量机 挠度修正 传感器 

分 类 号：U446[建筑科学—桥梁与隧道工程]