Short-Term Freeway Speed Profiling Based on Longitudinal Spatiotemporal Dynamics

Short-term traffic forecasting accuracy is related closely to the use of neighboring traffic information. Multivariate forecasting methods are becoming more popular because of their ability to capture both temporal and spatial evolvement in traffic. However, little attention has been given to quantify the effect of upstream and downstream traffic information, and the vast majority of published studies assume that the spatiotemporal relationship is specified in advance. Thus, the selection of surrounding traffic information as input parameters is somewhat arbitrary. To address that issue, this study investigated spatiotemporal relationships of speed series from consecutive segments under different traffic conditions by using the link speeds for nine segments extending over 12 mi on I-24 in Nashville, Tennessee. A prewhitened cross-correlation technique was proposed first to clarify the cross correlations between two speed series. The prewhitened cross-correlation function was performed on speed series for consecutive freeway segments for periods including the morning peak, midday off-peak, and evening peak. The analysis results showed that the correlations for consecutive segments were highly related to traffic conditions and that the effect of downstream traffic increased with the severity of congestion. Influences of upstream and downstream locations on current traffic were also found to be not symmetric in regard to the current site. The algorithm on properly choosing neighboring traffic information was proposed, and the lagged regression model with correctly identified input parameters (segments) outperformed others.