Practical Innovations and Experience in ABM Equilibration with
Practical Innovations and Experience in ABM Equilibration with Network Assignments Chrissy Bernardo, Peter Vovsha, Gaurav Vyas (WSP), Rebekah Anderson & Greg Giaimo (ODOT) 16 th TRB Planning Applications Conference
Background • 3 C ABM Concurrent ABM development for 3 major metropolitan areas in Ohio: 16 th TRB Planning Applications Conference 2
ABM Equilibration Land Use Network Level of Service • Global Feedback • Equilibrium between supply and demand • Convergence ensures uniqueness of model output for each scenario Accessibilities and O/D Measures Travel Paths by Mode O/D Travel Patterns by Mode Travel Demand Locations, Modes, and Tour Structure 16 th TRB Planning Applications Conference
Global Equilibration Benefits Costs • Uniqueness/replicability of solution • Equilibrium between supply and demand • More accurate/stable results Equilibrium criteria = compromise between benefits and costs 16 th TRB Planning Applications Conference 4
Strategic Feedback Options 4 -Step Models Averaging trip tables Averaging skims Averaging link volumes Averaging speeds Re-run mode choice only Skip transit assignment if no major impacts expected on transit modes Activity-Based Models All 4 -step options, plus: Re-run non-mandatory location choice / tour formation, time of day, & mode choice only Ramp-up demand (sampling) 16 th TRB Planning Applications Conference 5
Equilibration Strategies: MSA 16 th TRB Planning Applications Conference 6
MSA Dimensions: Hierarchy LOS Feedback: Zonal Skims + Trip Tables (for static assignment only) Link Speeds Link Flows 16 th TRB Planning Applications Conference Output / Drivers Direct Demand Feedback: Non-linear impact (path choice) Non-linear impact (BPR) 7
Convergence Analysis Ran a full set of global iterations for Lima, OH (3 C ABM) • 100% population • All steps run on each iteration • Transit Assignment (with feeding back dwell times based on ridership) • Accessibilities (impacts both long-term choices and activity participation) Used complete cold start (free flow speeds) Two MSA options: • MSA on link volumes only • MSA on link volumes and trip tables 16 th TRB Planning Applications Conference 8
MSA – Impact on Convergence Link-by-link RMSE - Assigned Volumes Link-by-link RMSE - Averaged Volumes 25 25 20 20 15 15 10 10 5 5 0 0 1 2 3 4 5 MSA over Volumes 6 7 8 MSA over Volumes & Trip Tables 9 10 1 2 3 4 5 MSA over Volumes 16 th TRB Planning Applications Conference 6 7 8 9 10 MSA over Volumes & Trip Tables 9
MSA – Impact on Convergence Trip Table RMSE - TAZ level 1. 60 1. 40 1. 20 1. 00 0. 80 0. 60 0. 40 0. 20 0. 00 1 2 MSA over Volumes 3 4 5 MSA over Volumes & Trip Tables 6 7 8 9 10 MSA over Trip Tables (Averaged) 16 th TRB Planning Applications Conference 10
Additional Equilibration Options Re-run selected ABM Steps • Potential options: • Re-run mode choice only (3 C) • Re-run non-mandatory location choice / tour formation, time of day, & mode choice only • Run time savings • Finer control over model sensitivity • Good for analyzing temporary projects or short-term impacts • Good for analyzing minor projects that are not expected to impact long-term choices • Projects that are temporary or very near-term may not affect long-term choices this option allows the model to evaluate immediate impacts • Re-running just mode choice & assignment does not capture how people may re-structure their activities (e. g. go shopping on the way home from work instead of making a separate tour, go out to dinner near workplace instead of near home, etc. ) 16 th TRB Planning Applications Conference 11
Additional Equilibration Options Ramp-up demand • Run time savings • Good for major network changes or future years • Sample successively larger proportions of the population to run through the ABM • Each iteration, a percentage of households are randomly sampled across the entire region to run through the ABM • Final iteration will always use 100% • Each household is weighted as necessary to represent the full population of travelers 20 to 30% sample provides enough spatial coverage to represent network LOS • Speeds up cold start convergence, allowing demand to respond to changed conditions in early iterations • Big changes to inputs may require more iterations to converge, this speeds up each global iteration 16 th TRB Planning Applications Conference 12
Run Time Savings – Estimates for Columbus Feedback Strategy Time Saved: 1 st of 3 Global Iterations Overall Time Saved: 3 Global Iterations Impact on Model Results MSA over Link Volumes -- -- MSA over Trip Tables -- Up to 1 global iteration 60% ABM run time 30% ABM run time Negligible Re-run Mode Choice Only -- 50% ABM run time Change in Modes Only Re-run Tour Formation + Only -- 40% ABM run time Changes in tour structure and non-mand. locations Ramp-up Demand 16 th TRB Planning Applications Conference Faster convergence None 13
Run Time Savings – Estimates for Columbus Feedback Strategy Impact on Model Results Recommended for: Example Project/Run MSA over Link Volumes Faster conv. All model runs Any MSA over Trip Tables Faster conv. All model runs Any Ramp-up Demand Negligible Major changes to inputs Future Year Re-run Mode Choice Only Change in Modes Capturing short-term impacts only New transit service (FTA approach) Re-run Tour Formation + Only Changes in tour structure and non-mand. locations Autonomous vehicles, major temporary network changes (e. g. highway closure, bridge collapse, rail line closure, etc. ) Capturing medium-term impacts (commuting patterns stay the same but other behavior may shift) 16 th TRB Planning Applications Conference 14
Contacts Chrissy Bernardo Technical Principal Systems Analysis Group Chrissy. Bernardo@wsp. com Peter Vovsha Assistant Vice President Systems Analysis Group Peter. Vovsha@wsp. com 16 th TRB Planning Applications Conference 15
- Slides: 15