EMPIRICAL MODE DECOMPOSITION AND CORRELATION PROPERTIES OF TRAFFIC FLUCTUATION

Abstract

This work describes the application of an important tool which can extract periodic information from the time-series of fluctuating traffic data. Compared to the traditional approach using FFT techniques, the nonlinear empirical mode decomposition (EMD) method has a number of advantages. This method is adaptive and therefore highly efficient at identifying embedded structures, even those with small amplitudes. Using this analysis, the traffic time series can be completely decomposed into five temporal modes including a 24-hour cycle, a 1-week cycle and a trend. Simultaneity, long-range correlations in traffic time series are investigated by detrended fluctuation analysis (DFA). In order to accurately capture the scaling exponents, EMD analysis is performed for DFA of the traffic records. The results of DFA for the data cleaned by subtracting the first intrinsic mode functions (IMF) are apparently improved, although the DFA curves are not entirely straight on the log-log plot.