Office of Operations
21st Century Operations Using 21st Century Technologies
Table of Contents

CHAPTER 7. STEP-BY-STEP APPROACH

The primary components introduced in chapter 5 could be used to develop a step-by-step calibration process so that the analytical/simulation models and tools to data that are reflective of the target future conditions.

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.

STEP 1

As the first step, prior to the commencement of any modeling, the study goals and objectives should be determined. At this stage, the granularity level (i.e., micro, meso, or macro) of the problem is specified. A list of performance measures consistent with the study goals should be defined. The relevant field data is another important element that should be identified. The following lists suggested data categories that could serve as input to the simulation tool:

  • Network coding data such as link length, number of lanes, lane connectivity, lane use restrictions, lane stripping, turn prohibitions, signal locations, and signal timing.
  • Simulation agent behavior data such as fundamental diagrams and driver’s reaction time.
  • Demand data such as link traffic flows and turning movements.
  • Incident data such as location, date, time of clearance, specification of the affected network segments, and severity.
  • Work zone data such as work zone activity type, intensity, location, date, time of occurrence, specification of the affected network segments, duration, restrictions during the work zone and activity.
  • Special event data such as the type and description of the event, location and area of impact, date, time of occurrence, and duration.
  • Weather data such as weather station location (longitude and latitude), date, time of weather record, visibility, precipitation type, precipitation intensity, and temperature.

A list of possible policies tied to the study goals that are being examined by the simulation tool should be provided. These policies would be used throughout the scenario generation.

STEP 2

After the first step, the models that would be used in the simulation should be selected. The models include microscopic models such as car-following, lane-changing, gap acceptance, queue discharge models; mesoscopic and macroscopic models such as fundamental diagrams at the network level and at the link level; and strategic models such as route choice, mode choice, and departure time choice models. Once the models are selected, the input data specified in step 1 could be transformed into variables that are used by the model. Next, the model parameters should be identified. Based on the problem specifications in step 1 and the available data, the parameters could be classified into the categories introduced in chapter 3 (parameters with the least level of uncertainty; parameters with some level of uncertainty; and parameters with deep uncertainty). Categorization of the parameters would help to determine the type of analysis that should be used.

At this stage, some parameters could be estimated from the data collected from the study area or could be prespecified using assumptions consistent with the problem. The remainder of the parameters could be extracted from the library of parameters. These libraries essentially provide a set of possible values for each model parameter. The parameter correlations should be considered when the parameter values are extracted from the libraries. The correlation between parameters is reflected in the joint probability distribution of the model parameters. Once all the parameters are extracted, the set of parameters that need to be calibrated should be determined. As a result, this step could be summarized into answers of the following questions:

  • Which models should be used in the simulation tool?
  • What are the variables and parameters of the selected model?
  • How should the available data be translated into the model variables?
  • What category does each model parameter belong to? (Type 1: parameters with the least level of uncertainty; Type 2: parameters with some level of uncertainty; and Type 3: parameters with deep uncertainty)
  • How could the available data be used to estimate some of the model parameters?
  • What are the parameters that should be extracted from the available libraries?
  • What are the possible set of values and joint probability distributions for the parameters that are taken from the libraries?
  • Which parameters need to be calibrated for the problem being studied?

STEP 3

In this step, the simulation agents should be generated. The parameters and models specified in the previous step characterize the behavior of the agents. The agents could be created by selecting parameter values from the joint probability distributions of the model parameters. Each set of values chosen for the parameters could be incorporated into the models to define the behavior of an agent.

STEP 4

Once the different agent types are generated in step 3, the various scenarios that would be analyzed by the simulation tool should be created. Scenarios are generated on the basis of the agents created in the previous step and the policies specified in the first step. Scenarios could be classified into three groups:

  • External event scenarios such as different weather conditions, incidents, and work zones;
  • Traffic control scenarios such as signal control, pricing, ramp metering, variable message signs, managed lanes; and
  • Travel demand scenarios such as market penetration rates of different agents, day-to-day demand variation, visitors demand, demand of special events, and closure of alternative modes.

The three classes of scenarios could be combined to generate new scenarios.

STEP 5

After fully characterizing the scenarios, the possibility of constructing a joint probability distribution for the scenarios should be examined. The joint distribution explains the relative importance of each scenario. Therefore, the joint distribution could be characterized based on:

  • The joint probability distribution of the model parameters.
  • Predictions for the realization of the scenarios under potential future conditions.
  • The relative importance of various scenarios according to an expert opinion.

If the joint probability distribution could be constructed, then the step 6-1 should be performed. Otherwise, in the presence of parameters with deep uncertainty, due to lack of information, the joint probability distribution for the scenario set could not be specified. In this case, step 6-2 should be conducted.

An important question to be answered is the number of scenarios that should be generated so that the simulation outputs represent the reality at an acceptable level. Recommendations regarding the number of scenarios are provided in appendix A.

STEP 6-1

Run each scenario in the simulation tool. Then, perform the scenario-based analysis to combine the simulation output using the probabilities assigned to the scenarios. As a result, the combined output is a weighted average of the simulation outputs with the weights equal to the scenario probabilities. The combined output could be changed to become a reasonable representation of the reality by changing the probabilities assigned to the scenarios. A possible approach for updating the probabilities is to Bayesian inference which is explained in appendix B.

The number of simulations for each scenario is an important question that should be addressed in this step. Depending on the available resources, recommendations are provided in appendix A.

Once the simulation output is calibrated by adjusting the probabilities of the analyzed scenarios, the probability distribution of each parameter and the scenario probabilities could be used to define a calibrated probability density function for each parameter.

STEP 6-2

Determine the desired robustness metric to use for combining the outputs of the simulations. Run each scenario in the simulation tool. Perform the robustness-based analysis on the simulation output based on the selected robustness metric. The number of simulations for each scenario is an important question that should be addressed in this step. epending on the available resources, recommendations are provided in appendix A.

According to the second transformation column in table 6, based on the selected robustness metric, a single scenario or a set of scenarios would be selected to perform the analysis. Once the simulation output under the robustness-based analysis is specified, the probability distribution of each model parameter and the selected scenario(s) could be used to define a proposed probability density function for each parameter.