Identification and Calibration of Site-Specific Stochastic Freeway Breakdown and Queue Discharge

The stochastic nature of freeway bottleneck breakdown and queue discharge is investigated through a comprehensive analysis of sensor data collected at bottleneck sites in the San Francisco Bay Area, California, and San Antonio, Texas. A new procedure was proposed to define the stochastic variation of the onset of freeway breakdown and of queue discharge capacity on the basis of time-indexed field data of speed–flow profiles. The former was developed as a function of average vehicle time headways preceding observed conditions when both speed was below and density was above locally defined congested flow thresholds. A full-year 15-min data series was used in the demonstration and testing of the procedure and yielded a high degree of statistical confidence in the resulting estimates of headway distribution parameters. The statistical analysis indicated that the probability function of freeway bottleneck prebreakdown headways followed a shifted lognormal distribution. In addition, a recursive queue discharge model was proposed for bottleneck flows under congested (queued) conditions. The proposed queue discharge model was a simple autocorrelated time series recursion that was seeded with the corresponding prebreakdown flow and dampens to the mean queue discharge rate. The proposed stochastic models are robust and accurate and represent a significant improvement in the understanding and modeling of freeway bottleneck flow. The models were implemented in the mesoscopic network model DYNASMART-P to test the effects of stochastic freeway capacity on sustained service rates and network performance.