During major highway construction, when lanes or entire highway sections must be temporarily closed, traffic managers would like to inform motorists of alternative routes around the construction site well in advance of the project location. This study develops a traffic diversion model to propose an optimum alternate route to drivers during a construction activity. The models and algorithms developed in this study assess a potential diversion route to optimize network performance while considering the drivers’ behaviors in following the proposed alternate route during a closure. A bilevel optimization model is proposed to minimize the total travel time of the affected network considering the link closure and a proposed alternate route for the travelers. A travelers’ route choice decision is modeled based on the user equilibrium traffic assignment, whereas a certain percentage of drivers are assumed to divert to the recommended alternate route. A sufficiently large subnetwork is selected, and a path selection method is proposed to reduce the computational effort required to optimize the model. A set of simulation experiments is conducted using the Tarrant County network in north Texas. The results show the ability of the model to improve the overall network performance during hypothetical closure scenarios.
This research integrated and evaluated emerging user data sources (Strava Metro, StreetLight, and hybrid docked/dockless bike share) of bicycle activity data with conventional “static” demand determinants (land use, built environment, sociodemographics) and measures (permanent and short-duration counts) to estimate annual average daily bicycle traffic (AADBT). We selected six locations (Boulder, Charlotte, Dallas, Portland, Bend, and Eugene) covering varied urban and suburban contexts and specified three sets of Poisson regression models: city-specific models, an Oregon pooled model, and all cities pooled. Static variables, Strava, and StreetLight complemented one another, with each additional data source tending to improve the model performance. Sites with lower volumes were more difficult to predict, with considerable error in even the best-performing models. City-specific models in general exhibited improved fit and prediction performance. Expected prediction error increased by a factor of about 1.4 when using Strava or StreetLight alone, but without static adjustment variables, to predict AADBT. Combining Strava plus StreetLight, but without static variables, increased error slightly less; by 1.3 times. We also found that transferring the model specifications from one year to the next without re-estimating the model parameters resulted in a 10% to 50% increase in error rates across models, so such transfer is not recommended. The findings from this study indicate that rather than replacing conventional bike data sources and count programs, old “small” data sources will likely be very important for big data sources like Strava and StreetLight to achieve their potential for predicting AADBT.