This page provides an overview of the highway skimming and assignment procedures. For more information on the inputs, outputs, and parameters of the ABM components, please see the Highway Inputs and Highway Networks and Skims pages.
Highway assignment is a multi-class static user equilibrium assignment with the following default user classes:
The solution to the traffic assignment problem is found using the Frank-Wolfe algorithm. The convergence criterion is a relative gap of 0.0001 achieved on three consecutive equilibrium iterations. A generalized cost function that includes travel time and toll cost is used to find the least cost paths at each user equilibrium iteration. The generalized cost is based on a $15 / hour peak value of time (VOT) and $12 / hour off-peak VOT. Distance / operating costs are not considered in the assignment procedure and the VOT is consistent across occupancy rates, i.e. the toll cost is not assumed to be shared amongst all passengers.
The default roadway capacity is input through a series of Capacity Lookup Tables. The Facility Type Class 2 (FTC2) and number of lane attributes of the link are used to determine the lookup table.
The default capacity is then adjusted for Freeway, Uninterrupted Roadway, and Toll Roads only depending on other attributes on the link, as shown in the following table.
SERPM 8 assigns 5 time periods independently. The time periods are identified in the following table
To Be Determined - May be simplified or removed
The highway assignment procedure is applied in an iterative fashion, where travel times are updated after each iteration to reflect congestion occurring on the network. These updates to travel time are based on a volume-delay function for each link. The existing volume-delay functions are the modified versions of the standard BPR functions with parameters varying by facility types, which will be reviewed and modified if necessary during validation. The free-flow time is based initially on the network data provided for each link and then updated in each iteration to represent the travel time resulting from the assigned traffic volumes from the last iteration.
Turn penalties are included in the trip assignment model to either prohibit certain turn movements or to penalize certain turn movements. These are included in the model by identifying specific turn movements by their node numbers, and then coding the penalty function that will apply to these turn movements.
The demand forecast procedure starts with seed highway travel time skim matrices, which are used to run the activity based model (ABM) and non-ABM models. The resulting vehicle trips are assigned to the highway network to obtain link level vehicle flows, V, and “congested” or loaded travel times. Since these travel times can be different from the times in the original seed skims, a new travel time is computed and the demand model is run again. This means of successive averaging (MSA) process ensures that the speed feedback iterations will converge by dampening the iteration to iteration changes. Link volumes from the current speed feedback iteration are averaged with the averaged results of previous iterations. Volumes are averaged by applying 1/iteration of the difference between the last MSA volumes with the current volumes. This process is repeated until equilibration is achieved.
The statistic used to establish that equilibrium between supply and demand has been achieved is the %RMSE of the AM and PM drive-alone travel time, computed over the full travel time matrix. The %RMSE is calculed in a demand weighted fashion, such that the OD travel time error is weighted proportionally to the number of trips. This statistic ensures that errors on frequently observed OD pairs are given more importance than errors on infrequently used OD pairs. When both the AM and PM weighted %RMSE are less than 1%, the model is considered converged and exists the speed feedback loop.
In addition to the default user classes, SERPM 8 also supports the optional segmentation of the advanced vehicle technology (AV) modes into separate user classes. The AV user classes are segmented by occupancy and are assigned without restriction to using toll facilities, i.e. are equivalent to the (pay) default classes. Note that the AV modes in mode choice are segmented into 'free' and 'pay', but all free AV modes are combined with the default user class for assignment. This is done because the assignment only treats AVs differently in access to HOT facilities, which 'free' modes are restricted from using. In summary, the optional user classes are:
For Future Mobility scenarios, there are two control switches that modify the restrictions: AV on HOT and TNC as AV. Note that the TNC as AV switch has no effect unless AV on HOT is also enabled. Also, note that the autonomous vehicle user classes are only created when AV on HOT is enabled. Otherwise, all vehicles are treated the same way in assignment and segmentation is not necessary. The restrictions are summarized in the following table: