针对由桥梁结构损伤和监测系统传感器故障引起的桥梁数据异常诊断问题,提出了一种基于主成分分析和超球面一类支持向量机的桥梁数据异常诊断方法。首先,对原始信号进行预处理和小波阈值去噪,然后,统计信号的时域、频域及自回归模型系数特征,利用主成分分析进行特征提取获得重要特征,最后,利用超球面一类支持向量机进行异常模式识别。通过江阴大桥伸缩缝相关监测数据表明,该方法可以较好识别和伸缩缝相关的数据异常,防止监浏系统误报漏报的发生。
The bridge data novelty diagnosis generated by the bridge structure damage or sensor fault, is a challenging problem for researchers in structural health monitoring area. A novel method has proposed which is based on a principal component analysis and hyperspherical oneclass support vector machine toward the problem. Firstly, the original signal is pre-processed and denoised by wavelet threshold. Secondly, for the filtered signal, the statistical features in time domain, frequency domain and the autoregressive coefficients are used to construct its fea- ture vector. Finally, the main feature vector generated from principal component analysis is presented to the hyperspherical one-class support vector machine for examination. The method is demonstrated on the monitoring data of an expansion joint in the Jiangyin bridge. It shows the accuracy and efficiency of the method for the identification of abnormal data.