Abstract: In order to handle the problem of fault in dynamic traffic data, an improved multi-scale principal component analysis (MSPCA) and a data modifying model were proposed. Firstly, using wavelet packet multi-scale decomposition, the individual variable was decomposed into approximation coefficients and detail coefficients of multiple scales and the corresponding principal component analysis models in various scale matrices were established. Using the model statistical magnitude as the threshold value, the comprehensive principal component analysis model was obtained by reconstructing wavelet coefficients and the fault data was separated. Secondly, using the data modifying model and correlation coefficients of each set of data calculated out by the time correlation and spatial correlation, the true value of the fault data was estimated. Finally, various simulation results were given.

Key words: traffic and transportation engineering, multi-scale principal component analysis (MSPCA), fault diagnosing, data modifying, wavelet packet

摘要： 针对动态交通数据的故障问题，提出了一种改进的多尺度主元分析（MSPCA）方法及数据修复模型。利用小波包多尺度分解将每个变量一次分解成逼近系数和多个尺度的细节系数，并在各个尺度矩阵建立相应的主元分析模型。以模型统计量控制限为阈值，对小波系数重构得到综合主元分析模型，并将故障数据分离出来。利用数据修复模型以及根据时间相关性和空间相关性计算出各组数据的相关系数，并估算出故障数据的真实值。最后给出了各种仿真结果。

关键词: 交通运输工程, 多尺度主元分析, 故障诊断, 数据修复, 小波包