Abstract
This paper proposes a novel filtering approach to scrutinize the characteristics of traffic flow dynamics measured in various time intervals during different times of day at different spatial locations. The proposed approach executes stepwise procedures to filter out periodic/quasi-periodic time series by Fourier power spectrum, equilibrium fixed points by the largest Lyapunov exponent, randomness by iterated function system (IFS) clumpiness maps, and followed by distinguishing chaoticity and stochasticity by correlation dimension. The dataset used in the numerical study is collected from the detector stations of Taiwan Freeway No. 1. The numerical study mainly analyzes the properties of dynamic traffic flows of four detection stations, sampled, respectively, in 20-second, one-minute, three-minute and nine-minute intervals, measured during midnight, morning, afternoon and evening periods. The results indicate that different nonlinear patterns, including fixed points, randomness, and chaotic-like phenomena may exist irregularly, contingent on the measured time intervals and observed times of day.
