Multivariate singular spectrum analysis for traffic time series

摘要

Multivariate singular spectrum analysis (MSSA) is a robust technique to analyze signals without any assumptions of the underlying system. It can decompose the original time series into a set of components, which are recognized as either a trend, periodic or quasi-periodic signal or residual noise. In this paper, this method is utilized to decompose multivariate traffic time series and then reconstruct them. We select proper parameters (window length and the number of eigenvalues) by defining an index and w-correlation matrix. Then we analyze the leading 12 reconstructed components for three detectors and observe that there exists a more detailed period in traffic system. By the reconstruction of original signals and analyzing the residual noise, the patterns for weekdays and weekends are different. (C) 2019 Elsevier B.V. All rights reserved.