Abstract: In the traffic flow time series, there are usually some random components which affect the accuracy of traffic character analysis and traffic forecasting owing to the breakdown of data collection equipment and communication. Thus, it is necessary to make the traffic flow raw data filtered. Wavelet analysis is widely used for signal de-nosing and has net been used in traffic research yet. This paper first introduces the occurrence of floating car’s noise and wavelet analysis, and then selects the filtering evaluation indexes. For the purpose of filtering effectiveness comparison, urban expressways of Beijing in different times are tested and data collected from RTMS as the real data. By the indexes comparison of similarity and mean square error, results and analysis of examples are provided to illustrate the advantage of this approach.

Key words: floating car, wavelet threshold de-noising, mean square error (MSE), correlation coefficient, signal-noise ratio (SNR)

摘要： 交通流时间序列中由于数据采集设备、通讯故障等原因会出现随机扰动，这些扰动对交通特性分析及交通预测的准确性会造成一定影响，因此有必要对交通流原始数据进行滤波处理。小波分析是一种应用比较广泛的滤波方法，但在交通方面应用较少。本文首先对浮动车噪声产生的原因进行分析，然后对小波分析方法进行介绍，确定了滤波评价指标。为便于对滤波效果比较，本文将RTMS系统采集的数据作为真值，选择北京市内多条快速路路段在不同时段进行实验，对比原始数据与滤波处理后的数据与RTMS的相似性及均方误差，表明滤波效果明显。

关键词: 浮动车, 小波阈值去噪, 均方误差, 相关系数, 信噪比