A study of relationships in traffic oscillation features based on field experiments

Despite numerous theoretical models, only limited field experiments have been conducted to investigate traffic oscillation propagation, and the relationships between traffic oscillation features (e.g., period, speed variation, spacing and headway) have not received quantitative analysis. This study conducts a set of field experiments designed to inspect such relationships. In these experiments, 12 vehicles equipped with high-resolution global positioning system (GPS) devices following one another on public roads, and the lead vehicle was asked to move with designed trajectory profiles incorporating various parameters. Measurements of five features are extracted from processing the field vehicle trajectory data with a time-domain method. Frequency analysis is also proposed with the Fourier transform method to verify the effectiveness of the features measured by the time-domain method. Compared to the frequency-domain method, the time-domain method yields more measurements with comparable quality and is more robust on trajectories with a small number of oscillation cycles. Then, a series of linear regression analyses reveal a number of new findings on the relationships between these features. For example, the time gap between two consecutive vehicles is negatively correlated with the speed standard deviation of the preceding vehicle and the initial speed of the following vehicle. It is also positively correlated with the average speed of the preceding vehicle and the initial spacing. The findings are helpful in constructing new microscopic traffic models better describing traffic oscillation dynamics. To illustrate this benefit, revised car following models are proposed to capture the relationship between time gap and other features. The simulation results show that the revised models yield better prediction accuracy (in range of 18% to 40% with the oscillation experiment dataset and in range of 30–63% with the stationary experiment dataset) than the classical models on reproducing real-world trajectories.