Long-term, regional travel demand models are essential tools used by planning organizations for resource management, project scheduling, and impact studies. In most cases, these tools are developed at a macroscopic level, including only the most basic information about the road networks’ geometry and traffic parameters. The development of the Green Line light rail corridor between the two Twin Cities downtown areas represented a modeling challenge for the Twin Cities Regional Planning Model (RPM) due to complicated road geometries and traffic controls. To explore this issue and a potentially better alternative analysis methodology as well as perform a before-after study, a different modeling approach based on traffic simulation and Dynamic Traffic Assignment (DTA) was developed and is presented in this report. Toward that end, a large-scale simulation was constructed to capture localized, high-resolution data and incorporate accurate transit and signal information while maintaining a wide, regional scope sufficient to capture longdistance travel and dynamic rerouting. Large-scale traffic simulation is by itself a very new capability allowed by recent advancements in traffic modeling as well as computer hardware and software. To that extent, very few examples of large-scale simulation are available and there are even fewer attempts to integrate DTA traffic simulation with a travel demand model. Even more, in this project a new approach for traffic simulation is utilized by using a simultaneous operation of two modeling resolutions, a microscopic level model for the core network around the two LRT Transitways and a mesoscopic level model for the rest of the metropolitan area. This Hybrid traffic simulation scheme provides for great detail in the project area without compromising route selection for trips originating and destined outside of it. The challenges were many for the implementation of this very new methodology and tool. Some of the challenges involved the development of the network geometry while maintaining a link to the RPM, the calibration of this large model, the designing of efficient experiments that minimize overall project duration and effort, and novel ways of visualizing the produced results since traditional approaches are not applicable to such a large network. This report describes the complete development of the hybrid model. As a primer, descriptions of similar projects are given along with broad background on varieties of simulation, culminating with the Aimsun hybrid simulation. The implementation of the geometric, traffic control, and vehicle demand components of the hybrid network is then described, including details regarding the two alternative models: one for the existing Green Line along University Ave., and one for the no-build alternative. Alongside the construction of the Aimsun model, interconnections to the Regional Planning Model in Cube Voyager were maintained. Throughout the development of the Aimsun hybrid model, calibration and validation efforts were undertaken to ensure the hybrid model would both align with the RPM and also produce representative results according to real network data. Calibration within this framework refers to adjustments made to network or vehicle parameters, which aimed to correct irregularities within simulated behavior. Calibration/Validation took place in steps, starting with the macroscopic level where an agreement with the RPM was established. This step was necessary since the hybrid and RPM models are designed over different software platforms each with its own abstractions and network coding ways. Four main issues were identified and corrected as part of the macroscopic calibration process: selfcentroid trips, unusual centroid configurations, volume-delay functions, and High Occupancy Vehicle behavior. The second step involved the calibration of the core microscopic parts of the model. This step ensured that reasonable traffic patterns were formed in the simulated arterials and freeways and ensured the correct operation of the 700+ traffic signals including LRT preemption and priority rules. Once the subarea networks were integrated into the larger regional hybrid model, manual calibration techniques became unusable due to the immense breadth of the network and the significant time cost to perform each iteration of simulation. As such, alternative calibration strategies were developed. These ‘blanket calibration’ techniques involved adjusting parameters for larger blocks of the network simultaneously, either by region or by targeting a particular subset of the network (e.g., all roads of a certain type, all nodes, etc.). The validation of the hybrid model was based on intersection turning counts provided by the cities of Minneapolis and St. Paul in various formats and from varying time periods and loop detector counts on freeway sections provided by MnDOT. During this process the research team had to also deal with numerous bugs still present in the software. Both project objectives were accomplished although the original plan and methodology had to be modified to accommodate the aforementioned challenges. The integration with the RPM Mode Choice step was accomplished and showed that it is feasible to upgrade the traditional static traffic assignment step with a Dynamic User Equilibrium based hybrid traffic simulation one. Although further improvements relating to the overall process are still needed, this proof-of-concept ensures that an integration with the new activity-based RPM is possible and potentially even more efficient and accurate. The comparison of the Green Line corridor with and without the LRT has produced credible results confirming that although the reduction in capacity of University Ave, does affect neighboring streets, the larger effect is absorbed by I-94 with only a marginal increase in travel times and reductions in speed. It is important to note that the model indicated the existence of instabilities on major intersections of the corridor where the selected traffic control plans play a big role in the system’s efficiency. Indicative results are presented both for the morning peak period (6:45 -7:30 AM) as well as the afternoon peak (3:00 to 4:00 PM), while a full result set for both peak periods is also available. The report concludes with a collection of lessons learned involving the development, calibration, and management of a large simulation model highlighting the challenges and suggesting an optimal course of action for future projects. (Author/publisher)
Samenvatting