Traffic prediction, data compression, abnormal data detection and missing data imputation: An integrated study based on the decomposition of traffic time series

摘要

This papers discusses the decomposition of road traffic time series and its benefits. The purposes of this paper are trifold. First, we provide an integrated framework for studying traffic prediction, data compression, abnormal data detection and missing data imputation problems, so that the relations between different problems can be revealed. In this part, we summarize several our works in this direction that had been finished in the last decade. Second, we discuss three most popular detrending methods: simple average detrending, principal component analysis (PCA) based detrending, as well as wavelet based detrending, and account for their intrinsic differences. Third, we present a new finding about trend modeling. We show that the detrending based prediction models previously designed for isolated sensor also work well for multiple sensors. Moreover, we define the so called short-term trend and explain why prediction accuracy can be improved at the points belonging to short trends, when the traffic information from multiple sensors is appropriately used. This new finding indicates that the trend modeling is not only a technique to specify the temporal pattern of traffic flow time series but is also related to the spatial relation of traffic flow time series.