-
Notifications
You must be signed in to change notification settings - Fork 21
CB Project Page
- Understand the impact of Mexican resident travel on the San Diego travel networks.
- Utilize the findings from the new 2019 Cross Border Survey and other data sources.
- Estimate border crossing wait time at POEs.
- Improve runtime performance and reduce software development and maintenance costs.
- Inform upcoming regional plan updates.
- Implement in the ActivitySim platform:
- To run in a reasonable amount of time.
- To reduce software development and maintenance costs in the long run.
- Integrated into the existing model system for ease-of-use.
- The model will be sensitivity tested to ensure reasonable sensitivity to typical policy, land use, and network scenarios.
- CT-RAMP code: code, uecs, properties file, inputs, parameters
- Underdevelopment ActivitySim code: code
As described in the RFP, the RSG team will begin the project by developing and delivering a Project Management Plan (PMP) to ensure the success of the project. The PMP will identify key team members, identify a schedule, and establish communication protocols. The communication protocol will include bi-weekly web meetings with SANDAG as well as periodic internal project meetings. During the bi-weekly meetings with SANDAG, we will discuss project progress, review draft deliverables, resolve technical issues, discuss upcoming work activities, and answer any questions that SANDAG staff have. RSG will use SANDAG’s confluence site for meeting notes, JIRA for issue tracking, and GitHub for version control, as stated in the RFP and has experience with SANDAG on all these platforms. Ad-hoc project meetings will be called as needed to communicate on special topics.
This will be one of the first new deployments of ActivitySim and as such we anticipate significant coordination with the ActivitySim consortium. RSG is uniquely positioned to lead this communication given our role as contractor for development of the ActivitySim software. At the outset of the project, we will fork the ActivitySim project into a SANDAG specific GitHub project, create a new tour-based model example and utilize this fork for model development. In spirit with the ActivitySim project, we propose the SANDAG specific project be public and any changes we make to the core ActivitySim code be submitted to the collaborative project for possible contribution. This will help ensure that the consortium is aware of our progress and make it easier for SANDAG to potentially use advances in the future. Other consortium members may wish to adopt the tour-based cross-border model structure for their special market modeling needs, or even as a faster, but less featured, model structure for resident travel.
RSG ensures quality in its technical products through a combination of involving experienced staff, having clearly defined staff roles and responsibilities, frequent communication with the client team, and the use of software management techniques such as GitHub, automated testing, and issue tracking. We will create a CBTM code base that is thoroughly tested and backed by a test system. As a result, this project will likely inform the ActivitySim consortium strategic development and contribution plan and eventual test system.
- Project Management Plan (PMP) Including Project Schedule
- Bi-weekly Project Status Meetings
- Bi-weekly Project Meeting Minutes Using SANDAG’s Confluence Site
- Two On-Site Meetings with Agendas, Presentation Materials and Minutes
- Monthly Progress Reports
- Ad-Hoc Project Meetings as required
RSG will develop a CBTM update plan that pivots off the existing design and makes use of new insights gleaned from the 2019 Cross Border Survey. Proposed Principal Joel Freedman developed the existing CBTM and will provide valuable direction on this task, along with input from SANDAG. Below we provide some initial thoughts on the model review and draft update plan:
- Some of the strengths of the existing model are its well-structured model steps - tour generation, time-of-day, station and destination choice, mode choice, and stop frequency, followed by trip purpose, time-of-day, destination choice, and mode choice- and its threaded implementation.
- Some of the weaknesses of the existing model are the lack of border wait time sensitivity to demand, lack of job competition for work travel, lack of network and zonal detail in Mexico, difficulty in maintaining the software, runtimes, error handling, and uncontrolled simulation variance.
- The existing SANDAG border studies are good resources to understand the cross-border traffic situation.
- The 2019 Cross Border Survey is a key resource in updating the CBTM. The survey will provide updated observed travel patterns of Mexican residents travelling in San Diego across all dimensions of travel – generation, distribution, mode, and time-of-day by market.
- The border wait time models for cars and pedestrians will take advantage of key characteristics such as operating lanes, cost, time-of-day, and demand. POE operations data will therefore be vital for estimating and validating the models.
- The current ActivitySim consortium software development plan includes support for model estimation mode, three zone systems (such as SANDAG’s Master Geographical Reference Areas (MGRAs), Traffic Analysis Zones (TAZs), and Transit Access Points (TAPs)), and transit virtual path building for calculating transit level of service between MGRAs via nearby TAPs. These functionalities will be incorporated into the CBTM update.
- A performant implementation of the model is one of the project requirements and it will be incorporated into the model update plan, as well as the Task 5 Model Implementation and Integration report. Based on prior RSG work for SANDAG to prototype multiple zone systems in the Activitysim framework, shorter runtimes are expected.
- RSG will refresh the existing model calibration with updated data and sensitivity tests for the model under different land use, POE service, and network scenarios. The prioritization of updates will be defined in cooperation with SANDAG. A draft model update plan will be shared with SANDAG and then finalized based on SANDAG comments before beginning additional tasks.
- Draft Model Update Plan
- Final Model Update Plan
Under Task 2, RSG will develop a detailed plan to build a set of consistent, comprehensive data for model estimation, calibration, validation, and application, which will then be processed under Task 3. In addition to the 2019 Cross Border Survey, POE operations data, Mexican resident data, transit boardings, and traffic counts are important data sets. The recently initiated 2020 Transit On-Board Survey will be a useful source of information on transit trips made by Mexican residents. We will process the data using Python, prepare a CBTM ActivitySim estimation mode database, and prepare a report on data development. RSG will include assumptions, data cleaning steps, any additional weighting/expansion that was required, and distributions of key variables such as time-of-day and mode in the data development report.
- Processed 2019 Cross Border Survey Data
- Python Programming and Scripts for Data Processing checked into SANDAG GitHub Repository
- Draft Data Development Report
- Final Data Development Report
It has been several years since the CBTM was estimated and so most of the models would benefit from re-estimation. The CB model is a tour-based model in which some of the components (tour and stop time-of-day choice, stop frequency, and stop purpose) are implemented as Monte Carlo draws from observed probability distributions. These models will be refreshed with data from the 2019 Cross Border Survey in Task 3. Other components (station choice, border crossing mode choice, tour and stop destination choice and trip mode choice) are logit models that must be statistically estimated from data. It will of course be important that the sample size available in the survey be sufficient to estimate models, for example, there must be enough transit riders in the survey to make mode choice estimation a useful exercise.
Model estimates will be summarized in spreadsheets and coefficients will be scrutinized to ensure that parameter estimates are significant, within reasonable ranges and appropriately signed. Model estimation results will be summarized in a model estimation report and implemented into ActivitySim input specification files (which are ActivitySim’s version of Utility Expression Calculator (UECs)). RSG plans to use the new ActivitySim estimation mode functionality to make estimation easier and to reduce model specification errors. All estimation files and scripts will be committed to the SANDAG GitHub repository.
- Draft Model Estimation Report
- Final Model Estimation Report
- Updated Utility Expression Calculators (UECs) Committed to SANDAG GitHub Repository
- Model Estimation Scripts Committed to SANDAG GitHub Repository
After model design and estimation, the model components will be implemented. RSG will implement each CBTM model component as a restartable data pipeline step. In cooperation with Task 2, RSG will develop and discuss implementation with SANDAG that meets the needs specified in the RFP: runtime, memory usage, exception handling, and the I/O data model. RSG will also identify procedures to be streamlined, simplified, and/or optimized and the possible tradeoffs in implementation that may be required to meet agency needs. An example tradeoff is faster model runtime versus increased memory needs in order to do more vectorized calculations at once. New model structures, such as shared station and destination choice with logsums will be implemented, which will require core ActivitySim software development.
The current base year CBTM processes ~115,000 tours and runs in 1 hours and 45 minutes. The re-write of the CT-RAMP MTC TM1 model into ActivitySim achieved about a 40% speedup in runtime. Some aspects of the CBTM are more complicated than TM1, such as the use of TAPs, while others are much less computationally intensive, for example no time windows and sampling from time-of-day distributions. As a result, we expect significant improvements in the ActivitySim-based CBTM runtime. The initial implementation of the model will be single processed. Multiprocessing will be added once the model calculations are verified.
We will manage the project backlog with SANDAG JIRA Issue Tracker and deliver the Python code with in-line API documentation (known as docstrings). Enriched in-line API documentation throughout the Python scripts will be provided to increase software readability. All code will be maintained in SANDAG’s GitHub fork. RSG will also integrate the CBTM with the rest of the model system, including the DOS-based run procedures, the reading of OMX skims, and the writing of trip matrices for assignment.
- Computer Programs Checked into SANDAG GitHub Repository with JIRA Ticket Descriptions
- A Draft of Uncalibrated CBTM Model with Reasonable Runtime Performance
RSG will calibrate the model to observed data. The goal of this process is to have a reasonably well-calibrated and validated travel model, suitable for sensitivity testing. RSG suggests that the following calibration summaries be performed, subject to refinement in Task 2. Note that certain model components (tour generation, tour and stop time of day, stop frequency, stop purpose) are simulated from probability distributions; therefore calibration is not necessary. However these model components will be summarized to ensure that the models are implemented correctly.
- Border crossing station and mode choice - Tours by border crossing mode, station, and market segment (including Sentri\non-sentri).
- Tour destination choice - Station to primary destination average distance and tour length frequency distribution and tours by origin and primary destination district
- Tour mode choice - Tours by tour purpose and mode Trip destination choice - Average out-of-direction distance and trip length frequency distribution and trips by origin and primary destination district
- Trip mode choice - Trips by trip purpose and trip mode, transit technology
RSG will validate the CBTM network assignments using SANDAG provided validation templates that make use of traffic counts and transit boardings. Model validation is an iterative task with model calibration; often the network assignment process reveals issues with underlying travel model calibration or specification that must be addressed in order to improve assignment results. A final technical report will conclude this task.
- A calibrated and validated CBTM updated model
- Draft Model Calibration and Validation Report
- Final Model Calibration and Validation Report
- Utility expression calculators (UECs) with updated parameters and constants
- Computer programs and scripts for model calibration/validation committed to SANDAG GitHub
The sensitivity tests will be identified in the Task 2 work plan and the results will be documented in a technical memo. The recent cross-border model sensitivity tests performed by SANDAG that relied upon the Travel Model Improvement Program Exploratory Modeling and Analysis Toolkit (TMIP-EMAT) would be a useful baseline for sensitivity testing for the new model. In addition, SANDAG staff may perform other sensitivity tests based on the business needs. The Proposer shall work with SANDAG staff and provide guidance on if the test plan is appropriate to reveal the sensitivities. RSG will correct any model issues that are discovered from the sensitivity tests.
- 5 sensitivity tests at minimum and guidance on the other sensitivity tests conducted by SANDAG Staff.
- Draft Report on Sensitivity Tests
- Final Report on Sensitivity Tests
RSG will develop a model development report that includes the topics requested in the RFP, which generally follow the work tasks: project objectives, data sources and data processing procedures, model estimation, model calibration and validation, software improvements, and sensitivity test results. The final report will be a compilation of the task reports developed during the project. To save on project resources, RSG suggests delivering a draft task report for each task and responding to comments in a Draft Final Report rather than responding to comments in a final task report. The draft final report also serves to provide an opportunity to respond to comments on the final report before it is published. The project resources saved by not producing final task reports for each task can be used for other tasks, such as additional implementation improvements.
- Draft Final Report
- Final Report
RSG will develop the initial model implementation on RSG Windows servers. Once in good working order, RSG will remotely deploy the updated model on SANDAG Windows servers. Once accepted by SANDAG, RSG will provide as-needed support to SANDAG staff for up to sixty hours addressing all aspects of running the updated CBTM. Support will be available through email, teleconference or in person communications. Support may include investigating or resolving model system run crashes, addressing counter-intuitive results, or any other requests within project budget.
- Deployed base year model on SANDAG servers
- Technical support up to 60 hours for SANDAG staff
- Completion by end 2021