CHAPTER 8. CONCLUSIONS

Note: Unless accompanied by a citation to statute or regulations, the practices, methodologies, and specifications discussed below are not required under Federal law or regulations.

Models incorporated in simulation tools typically contain mathematical relations. The relations are specified by a set of parameters. Calibration entails the process of estimating values of the model parameters to provide the best representation of the context and study area. Building on previous studies on calibration of simulation tools, this project provides a framework to develop models that are capable to produce realistic predictions of the available problem under potential future conditions and contemplated policy/operational interventions. The motivation for this study was to relax the assumption that the model parameters values remain constant over time. Parameter values may be different under future conditions than they are under the conditions for which the model was calibrated.

In recent years, some research studies on the calibration of traffic analysis tools have provided more information about the potential effectiveness of certain heuristic methods and fitness functions.

However, they have not provided robust models that are capable of correctly representing the system behavior for future conditions. On the other hand, other studies attempted to address this challenge by proposing an overall methodology or framework for modeling future conditions. One of the commonly used methodologies is scenario-based simulation. A scenario is defined by a set of operational conditions, interventions, as well as characteristics of the general activity system and associated technologies. In order to improve the accuracy of models in forecasting network behavior under various situations, a set of scenarios could be generated that reflect the expected future conditions of the network. Furthermore, reasonable probabilistic distributions for the scenarios would be used in order to combine the effect of the scenarios. As a result, scenarios play a significant role in depicting a realistic view of the future status of the system. Therefore, the scenario-based simulation methodology constitutes a major component in this study.

Other components of the proposed framework were defined based on other possible sources of error that were identified in previous studies. Besides the scenario-based simulation, parameter correlation, vehicle (and other simulation agents) trajectories, robustness-based simulation, and local density are other major components of the framework.

Recognizing the correlation among model parameters could produce a more reliable model by more accurately capturing the behavior of actors in the transportation network. To preserve the correlation among parameters, there is a need for developing library of parameters. These libraries could be used in different studies as a source of extracting parameter values for different models.

Besides the significant value of parameter libraries, trajectories provide the most complete description of the simulation agent’s behavior and the system state by retaining the ability to extract stochastic properties of both individual behaviors and performance metrics. The library of parameters and trajectories could provide a comprehensive understanding of the context and study area. Local density recognizes the strong correlation between traffic congestion and driver behavior.

It is defined as the density of traffic in the local vicinity of a vehicle perceived by the drive. Based on the local density, fundamental relationships between driver behaviors can be developed under existing conditions, and then re-used in future models having different densities on each segment. These relationships could be stored as a set of models and parameters in the parameter libraries.

Vehicle trajectories are still not regularly available to traffic analysts, though the situation is rapidly changing with greater willingness by system integrators and data vendors to share this information (albeit with all kinds of limitations on use). However, greater deployment of connected vehicle systems promises to dramatically increase the availability and accuracy of this type of data. Furthermore, the data collected by the sensors of connected vehicles could be used to identify trajectories of other actors (manually-driven vehicles, bicyclists, and pedestrians) in the transportation environment. A complete knowledge of trajectories could be used to specify the feasible set of values for parameters of behavioral models. Such a complete knowledge can support development of model and parameter libraries that could be used for simulation practices.

While the scenario-based simulation, as one of the main concepts in this study, helps to calibrate model parameters when a fully specified set of scenarios are imported to the simulation tool, the notion of robustness becomes helpful when the uncertainty level of parameters and scenarios increases. Since future conditions is naturally governed by uncertainty especially due to lack of data, a sensitivity analysis that accounts for all possibilities could be helpful in developing a reliable model. The robustness-based simulation incorporated into the framework is provides a systematic approach toward such sensitivity analysis.

The proposed framework is predicated on four key notions:

1. That the models themselves are responsive to the features of the future scenario (i.e., that the descriptors that specify a particular scenario be included in the model specification).
2. The definition of a library of model parameters corresponding to different types of agentsunder varying conditions.
3. The role of scenarios in both specifying future conditions of interest as well as triggering certain ranges of parameter values for those scenarios.
4. The potential for robustness analysis in situations where uncertainty about future scenarios or conditions is large.

Based on the proposed framework and its main components, a step-by-step approach is devised. Follow-on studies could apply the framework and the information to develop robust models that present meaningful and reliable predictions of the system performance under future conditions and contemplated policy/operational interventions. Although the case studies of this report show the application of the proposed calibration framework to a microsimulation problem, the framework could be used for simulations with any granularity level (i.e., micro, meso, or macro).