Understanding Simulation Variation in the BellevueKirkland-Redmond AB Model

Understanding Simulation Variation in the BellevueKirkland-Redmond AB Model

Understanding Simulation Variation in the BellevueKirkland-Redmond AB Model TRB Planning Applications, Raleigh, North Carolina May 16, 2017 Purpose To develop an improved sampling plan for the new Bellevue, Kirkland, Redmond (BKR) activity-based model (BKRCast) To understand stochastic variation in the model To select a sampling strategy for the model that increases stability and reduces runtimes 2 RSG BKRCast Model BKRCast is based on PSRC's SoundCast AB model Both use DaySim for demand and EMME for supply Three key differences: Smaller zones in BKR, larger zones outside Fewer time periods for revised networks Over/under-sampling of households by TAZ to increase model stability and decrease run time Currently re-calibrating/validating 3 RSG BKRCast Model Region

BKR part of PSRC Region Persons BKR (300k) Seattle (650k) Rest of King (1,050k) Pierce (800k) Snohomish (720k) Kitsap (250k) Total 3,770k persons 4 RSG Sampling Sampling Traditional ABMs specify a global household sample rate Typically used for reducing runtimes DaySim reads synthetic households with an expansion weight Usually set to 1

We can modify these weights to over/under sample the population by user-specified market 6 RSG TAZ Level Sampling BKRCast needed sampling by region (by TAZ) TAZ level sampling steps: Run the population synthesizer as usual Specify a sample rate for each TAZ Assign HHs into HH size and income bins by TAZ Sample HHs by bin to match TAZ level sample rates Set the expansion weight of each record Run DaySim as usual Average trip matrices before assignment 7 RSG TAZ Level Sampling Example Oversample BKR zones @ 400% Replicate a synthetic household four times

Set the weight to 0.25 Undersample zones outside BKR @ 25% Select 1 in 4 synthetic households Set the weight to 4 Sampling controlled by HH size and income 8 RSG Developing a Sampling Strategy Sampling Plans Tested Scenario BKR Seattle Rest of King Pierce Snohomish Kitsap TOTAL Outside BKR Run Time (mins) 1 100 100 100

100 100 100 100% 100% 94 2 100 25 10 5 5 5 17% 10% 19 3 100 50 20 10 10 10 27% 21% 26 4 100

100 40 20 20 20 46% 41% 44 5 200 25 10 5 5 5 25% 10% 28 6 200 50 20 10 10 10 35% 21% 36

7 200 100 40 20 20 20 54% 41% 52 8 200 100 100 100 100 100 108% 100% 105 9 400 25 10 5 5 5 41% 10%

43 10 400 50 20 10 10 10 51% 21% 52 11 400 100 40 20 20 20 70% 41% 68 12 400 100 100 100 100 100

124% 100% 121 13 600 100 40 20 20 20 86% 41% 83 14 800 100 40 20 20 20 101% 41% 102 15 1600 50 20 10

10 10 146% 21% 157 BKR sample rates 100%, 200%, 400%, 600%, 800%, and 1600% Outside BKR sampling rates 10%, 21%, 41%, and 100% 10 RSG For Each Sampling Plan Run 10 different random seed model runs Only run the demand model All inputs fixed including shadow prices Except for synthetic population, which varies The nave 100% sample run takes 90 min In all, 150 DaySim runs 11 RSG How We Measured Variation / Stability Variation = Std. Dev. / Mean Measures Geographic Levels

Cars per households Total trips Work trips per households Avg. commute distance Trips per tour Transit share Active transportation (walk + bike) share Avg. trip distance Avg. trip time AM trip share PM trip share Regional

Within BKR and outside BKR Sub-region (B, K, and R) District TAZ 12 RSG Verifying Convergence Run 20 seed runs for two sample plans BKR 100%, outside 10% BKR 400%, outside 10% Calculate cumulative average mean for different measures, such as total II, XX trips Shows convergence with more runs, but very close to converged already 13 RSG Preferred Sampling Plans Preferred Plans 2 1

Preferred Plan 1: BKR 400% Outside 21% Regional 51% Preferred Plan 2: BKR 200% Outside 21% Regional 35% 15 RSG Preferred Sampling Plan Fixed Analysis Outside BKR fixed at 21% with different within BKR sample rates Variation within BKR goes down with higher sample rate Within BKR fixed at 400% with different outside BKR sample rates Generally, less variation outside BKR with higher sample rate 16 RSG

Preferred Sampling Plan Fixed Analysis Outside BKR fixed at 21% with different within BKR sample rates Variation within BKR goes down with higher sample rate Within BKR fixed at 400% with different outside BKR sample rates Generally, less variation outside BKR with higher sample rate 17 RSG Preferred Sampling Plan Fixed Analysis Outside BKR fixed at 21% with different within BKR sample rates Variation within BKR goes down with higher sample rate Within BKR fixed at 400% with different outside BKR sample rates Generally, less variation outside BKR with higher

sample rate 18 RSG Total Trips Within BKR by Geography More variation with smaller geographies 19 RSG Variation of Total Trips by BKR Districts and Sample Rate Std. Dev per Mean BKR 100%, Outside 20% BKR 200%, Outside 20% BKR 400%, Outside 20% 20 RSG

Conclusions & Next Steps All solutions quite stable Selected preferred sampling plans that are 1/2 and 1/3 runtime compared to the nave 100% sample and very stable Scenario BKR Seattle Rest of King Pierce Snohomish Kitsap TOTAL Outside BKR Run Time (mins) 1 100 100 100 100 100 100 100% 100% 94 2 100 25 10 5

5 5 17% 10% 19 3 100 50 20 10 10 10 27% 21% 26 4 100 100 40 20 20 20 46% 41% 44 5 200 25

10 5 5 5 25% 10% 28 6 200 50 20 10 10 10 35% 21% 36 7 200 100 40 20 20 20 54% 41% 52 8

200 100 100 100 100 100 108% 100% 105 Preferred Sampling Plan 2: BKR 200% Outside 21% Regional 35% 9 400 25 10 5 5 5 41% 10% 43 10 400 50 20 10

10 10 51% 21% 52 11 400 100 40 20 20 20 70% 41% 68 12 400 100 100 100 100 100 124% 100% 121 13 600 100

40 20 20 20 86% 41% 83 14 800 100 40 20 20 20 101% 41% 102 15 1600 50 20 10 10 10 146% 21% 157 Preferred Sampling Plan 1:

BKR 400% Outside 21% Regional 51% Schedule permitting, investigate lower sample rates and destination choice alternative sample sizes for stability vs. runtime 21 RSG Contacts www.rsginc.com Ben Stabler [email protected] Nagendra Dhakar [email protected] Mark Bradley [email protected] https://github.com/RSGInc/BKRCast

Recently Viewed Presentations

  • Travelling huge distances - how to do it the green way How to ...

    Travelling huge distances - how to do it the green way How to ...

    www.geog.biz!!!! Do the revision on. www.geog.biz!!!! If you were travelling across the USA how would you do it the . GREEN . way? How would you keep your . carbon footprint . small (this is where you try to cut...
  • Use the method of cylindrical shells to find the volume of ...

    Use the method of cylindrical shells to find the volume of ...

    Arial Times New Roman Default Design Microsoft Graph Chart Use the method of cylindrical shells to find the volume of solid obtained by rotating the region bounded by the given curves about the x-axis {image} Set up, but do not...
  • Eugenics - PC\|MAC

    Eugenics - PC\|MAC

    Eugenics. Eugenics ("good birth") term coined by British scientist ; movement dedicated to improving the human race by better breeding . Eugenics is stemmed from Italian criminalist César Lombroso who said entire groups of people were born to be criminals
  • Employment Portion - Oklahoma State University-Stillwater

    Employment Portion - Oklahoma State University-Stillwater

    Canaan, Virginia C. Police Sergeant (63-6891) Police Officer I (PO I) Hiring Range: $10.69 to $11.76 hr.; $22,235 to $24,461 annually ... State board of Tests Intoxilyzer Certification. Same as SGR. Three yrs minimum service as SGT. Completion of CLEET...
  • Simple Banner - ERNET

    Simple Banner - ERNET

    'Simple Banner' is a clean template . Is perfect for personal or business and corporate use. More about 'Simple Banner'
  • Writing Workshop

    Writing Workshop

    Objectives. Today's focus: Showcase and walk participants through the MDPT support materials, Help educators develop a deep understanding of similarities and differences between extended, more "polished" writing vs. shorter "on-demand" writing, and why both are important for students, and
  • Literary Devices and Techniques

    Literary Devices and Techniques

    DEAD METAPHOR - An overused metaphor that has gained a new meaning. IMPLIED METAPHOR - A more subtle use of a metaphor. MIXED METAPHOR - Two or more metaphors used together that appear illogical. EXTENDED METAPHOR - Metaphor that continues...
  • Chapter Ten - Faculty Server Contact

    Chapter Ten - Faculty Server Contact

    Masculinities and Crime: Male Social Support Theory. DeKeseredy and Schwartz connect the macro level (a system of social patriarchy) with the micro level (an individual's decision to use violence against a specific woman) with the male peer group as the...