Evaluating Fluctuations in Urban Traffic Data and Modelling Their Impacts

Abstract

Predictability is important in decision making in many fields, including transport. The fluctuations of time-varying and short-term traffic processes pose critical challenges for real-time traffic predictions, which are the backbone of the intelligent transportation systems (ITS). The success of different prediction techniques depends on the structure of the phenomenon, in particular whether it is easily predictable or complex. Therefore, understanding the various aspects leading to high/low complexity in traffic time series data is critical in evaluating the performance of urban traffic systems. This thesis presents the application of the Hurst exponent metric from Fractal theory in quantifying fluctuations in different micro and macroscopic parameters of urban traffic. Data from three distinct sources and geographical locations are used for the analysis. i) Vehicle trajectory data collected every 0.5sec from an arterial road in India and a freeway in the USA are utilised to evaluate fluctuations in lateral movements and speeds of thousands of vehicles. The effect of the average lateral position and vehicle type on fluctuations is also examined. ii) Loop detector data collected every 30sec from monitor sites on an urban motorway in Australia are used to quantify fluctuations in speed and flow. The effect of the time of the day, weekend, proximity to ramps, and the presence of a horizontal curve, on fluctuations is explored using the analysis of variances. Further, latent class and random parameters Tobit models are estimated to predict the effect of the Hurst exponent (long-range dependence) of speed on crash rates at motorway sites. iii) Traffic count data collected every 5min at signalised intersections in Sydney, Australia are studied to understand the patterns of predictability. The effect of the day of the week, public holidays, special events, weather, etc. on predictability is discussed using a random effects linear regression model. This thesis is data-driven; the empirical results suggest the need to evaluate fluctuations in traffic time series data before using different prediction and simulation techniques. Furthermore, this thesis could lead to many applications of fractal analysis on highways and urban traffic.