SF-CHAMP Basics: Version 4.3 AKA Fury
Download as PPTX, PDF2 likes53,548 views
SF-CHAMP is an activity-based transportation demand model that predicts the daily travel of individuals in San Francisco. It uses census data, surveys, and traffic counts to simulate individual travel choices like tour generation, destination choice, mode choice, and more. The model has high spatial resolution at the traffic analysis zone and transit stop level. It has been validated against observed volumes and transit boardings. Recent updates include dynamic traffic assignment, bike route choice modeling, and accounting for transit crowding.
More Related Content
Similar to SF-CHAMP Basics: Version 4.3 AKA Fury (20)
SF-CHAMP Basics: Version 4.3 AKA Fury
- 1. SF-CHAMP Basics Version 4.3 AKA Fury Lisa Zorn Presentation to the SFMTA Intern/New Staff Seminar August 3rd, 2012
- 2. Activity-Based Modeling SF-CHAMP is a tool which predicts activities, locations, and travel time for every individual traveler in San Francisco Based on Census Data and local surveys MTC 1996 and 2000 Home Interview Activity Diaries Census 2000, 2010, and ACS Muni 2004/2005 Onboard passenger survey 2007 Stated Preference Survey: Pricing Muni “APC” count data Traffic Counts (MTC/Caltrans/SFMTA Simulation of every Bay Area resident’s daily choices (and visitors too) Trucks and “external” trips borrowed from MTC Model SF-CHAMP Model Basics 2
- 3. Activities Grouped into “Tours” HOME BASED TOUR 7 = Tour A tour is an entire chain of DESTINATION SECONDARY = Trip HOME-BASED trips: from your primary origin, TOUR Number indicates trip order to all of your destinations, 6 and then back again. HOME 1 Primary destinations PRIMARY TOUR: INTERMEDIATE vs. 5 STOP ON Home-based WAY TO WORK Intermediate stops Work 2 WORK Consequences of choices: Do you have a car available? Did you leave the car at home? 3 4 WORK-BASED Do you have a complicated day? SUB-TOUR WORK-BASED DESTINATION SF-CHAMP Model Basics 3
- 4. SF-CHained Activity Modeling Process Population Synthesis • Land Use input • Census Data Household + demographics such as ages, income, hh sizes, workers SAN FRANCISCO COUNTY TRANSPORTATION AUTHORITY 4
- 5. SF-CHained Activity Modeling Process Workplace Location Choice • Land Use input • Census (CTPP) • Modes, costs, distances Workplace Destination SAN FRANCISCO COUNTY TRANSPORTATION AUTHORITY 5
- 6. SF-CHained Activity Modeling Process Vehicle Availability • Accessibility of home & work • Accessibility between them Household Vehicles SAN FRANCISCO COUNTY TRANSPORTATION AUTHORITY 6
- 7. SF-CHained Activity Modeling Process Tour Generation • Accessibility of home & work • Accessibility between them • Demographics Tour pattern for the day SAN FRANCISCO COUNTY TRANSPORTATION AUTHORITY 7
- 8. SF-CHained Activity Modeling Process Tour Destination Choice • Initial tour schedule • Accessibility Tour Destinations SAN FRANCISCO COUNTY TRANSPORTATION AUTHORITY 8
- 9. SF-CHained Activity Modeling Process Tour Mode Choice • Accessibility to destinations for that time of day by mode Tour Modes SAN FRANCISCO COUNTY TRANSPORTATION AUTHORITY 9
- 10. SF-CHained Activity Modeling Process Intermediate Stop Choice • Tour pattern requirements • Accessibility of potential stops given tour mode Intermediate Stops SAN FRANCISCO COUNTY TRANSPORTATION AUTHORITY 10
- 11. SF-CHained Activity Modeling Process Trip Mode Choice • Cost, Travel Time • Demographics • Tour Mode Trip Modes SAN FRANCISCO COUNTY TRANSPORTATION AUTHORITY 11
- 12. Spatial Detail - Transit Every transit stop Every transit line Every street SF-CHAMP Model Basics 12
- 13. Sample Auto Volume Plot SF-CHAMP Model Basics 13
- 14. Roadway Calibration Data Calibrated BPR functions using speed and volume sensors for base year SF-CHAMP Model Basics 14
- 15. Validation: Volumes And Boardings Calibrated using 2000 base year data Validated to 2005 counts and boardings, and 2006 speeds E s tim a te d vs . O b s e rve d R o a d V o lu m e s 160000 Traffic Volumes Muni Daily Boardings by Route Daily Boardings by Route Estimated vs. Observed Estimated vs. Observed 140000 50,000 45,000 120000 40,000 100000 35,000 Estimated Daily Boardings 30,000 80000 25,000 60000 20,000 15,000 40000 10,000 20000 5,000 0 0 0 20000 40000 60000 80000 100000 120000 140000 160000 0 5,000 10,000 15,000 20,000 25,000 30,000 35,000 40,000 45,000 50,000 Observed Daily Boardings D a ily O b s e rv e d V o lu m e SF-CHAMP Model Basics 15
- 16. Van Ness BRT – Transit Line Validation Validated using 2007APC Data SF-CHAMP Model Basics 16
- 17. Newly Added Data • New estimations (BATS 2000): • Mode choice • Auto Availability • New calibration (ACS) • Workplace Location Choice • Auto Availability • Tour mode choice • New validation • 2010 APCs and Ridership • Recent Traffic Counts SAN FRANCISCO COUNTY TRANSPORTATION AUTHORITY 17
- 18. Land Use Inputs Households, Jobs, Households,Jobs, & ABAG Population & Population Countywide Totals SF Planning Dept. ABAG Households & Jobs Households& Jobs SF TAZs (Plan B) Non-SF TAZs ABAG/MTC All TAZs Income & Age & Age TAZ Level Land Use for Bay Area SF-CHAMP Model Basics 18
- 19. Data Storage • HDF5: • easily viewable in ViTables, HDFview • not as heavy a relational database • easily scriptable in Python, R • compressed • Free • Use for: • Trip diaries : • easy to write scripts at the end to mine the data • Skim matrices: • removes our Cube matrix 32-bit dependencies SAN FRANCISCO COUNTY TRANSPORTATION AUTHORITY 19
- 20. Code Base • Primarily C++ with Boost library • Secondary is Python with 64-bit numpy • 64-bit operations • Can bring more skims into mode choice • More distributed processes SAN FRANCISCO COUNTY TRANSPORTATION AUTHORITY 20
- 21. Spatial Detail – Analysis Zones • Trips are aggregated into “zones” • 981 zones in San Francisco • 1,275 in other Bay Area counties SF-CHAMP Model Basics 21
- 22. New Component – Bike Route Choice • Data: • CycleTracks smartphone app • Methodology: • Choice set generation: doubly stochastic method • Path size logit estimation
- 23. New Component – Bike Route Choice Attribute Coef. SE t-stat. p-val. Length (mi) --1.05 0.09 --11.80 0.00 Turns per mile --0.21 0.02 --12.15 0.00 Prop. wrong way --13.30 0.67 --19.87 0.00 Prop. bike paths 1.89 0.31 6.17 0.00 Prop. bike lanes 2.15 0.12 17.69 0.00 Cycling freq. < several per wk. 1.85 0.04 44.94 0.00 Prop. bike routes 0.35 0.11 3.14 0.00 Avg. up-slope (ft/100ft) --0.50 0.08 --6.35 0.00 Female --0.96 0.22 --4.34 0.00 Commute --0.90 0.11 --8.21 0.00 Log(path size) 1.07 0.04 26.38 0.00 SAN FRANCISCO COUNTY TRANSPORTATION AUTHORITY 23
- 24. New Component – Bicycle Assignment Bikes / hour 0 20 180 360 SAN FRANCISCO COUNTY TRANSPORTATION AUTHORITY 24
- 25. Bike Accessibility: From 4th and King SAN FRANCISCO COUNTY TRANSPORTATION AUTHORITY 25
- 26. Bike Logsums: From 4th and King Effect of Bike Plan Build SAN FRANCISCO COUNTY TRANSPORTATION AUTHORITY 26
- 27. New Component – Network-based Pedestrian LOS Forecastable, continuous pedestrian utility along walk path SAN FRANCISCO COUNTY TRANSPORTATION AUTHORITY 27
- 28. New Pedestrian and Transit Environment Factors • Attributes Empirically Estimated: • Hills (rise) • Indirectness • Population and Employment Densities (logs) • Street capacity
- 29. Example: Walking to SFCTA Work Purpose SAN FRANCISCO COUNTY TRANSPORTATION AUTHORITY 29
- 30. Transit Walk Access Links: Perceived Weight Walk-Local-Walk, Destination Ferry Building SAN FRANCISCO COUNTY TRANSPORTATION AUTHORITY 30
- 31. Transit Assignment with Crowding • Motivation: • Transit is crowded today and expected to get worse in SF • Failure to represent transit capacity leads to: • Unrealistically high forecasts • Poor line-level validation • No relationship between capacity projects and effectiveness measures such as: mode share, emissions, travel time SAN FRANCISCO COUNTY TRANSPORTATION AUTHORITY 31
- 32. Transit Assignment with Crowding In real-life, if a line is crowded one can: A. Wait for a vehicle with room (+ wait time) Same Trip: B. Walk to an earlier stop (+walk time, +ivt) Change Route C. Walk to another line (+walk, +ivt/wait) D. Switch modes E. Switch time periods Change Travel Plans: Occurs in core activity model F. Change destinations G. Not make trip SAN FRANCISCO COUNTY TRANSPORTATION AUTHORITY 32
- 33. Transit Assignment with Crowding Route changing algorithm implemented: • Boarding availability is f(crowding level) • If boarding is prevented, transit skim searches for next best route by: • Walking to an earlier stop • Taking a slower line that arrives at that stop • Walking to a different line • Iterations averaged until reach a stable solution: • Skim is representation of average of walk times and in-vehicle times SAN FRANCISCO COUNTY TRANSPORTATION AUTHORITY 33
- 34. Transit Assignment with Crowding • Dwell time on transit now a f(boardings, alightings) • Estimated based on APC data Dwell Time Articulated (sec) = 7.35 + (3.01 × boardings) + (2.04 × alightings) Dwell Time Standard (sec) = 4.89 + (3.72 × boardings) + (2.11 × alightings) SAN FRANCISCO COUNTY TRANSPORTATION AUTHORITY 34
- 35. Transit Assignment with Crowding Reflection of crowding in change of travel plans: • OD pairs with crowding will result in sub-optimal skims • Travel time skims flow up through the model chain via logsums to affect mode choice, time of day choice, destination choice, and tour generation SAN FRANCISCO COUNTY TRANSPORTATION AUTHORITY 35
- 36. SF Citywide Dynamic Traffic Assignment Why DTA? It’s a Better Representation of Reality! • Queues exist and are considered • Finer network detail that includes: • Transit vehicles that interact with cars • Intersection control • Signal timings • Intersection geometry SAN FRANCISCO COUNTY TRANSPORTATION AUTHORITY
- 37. Why DTA? Because we shouldn't just discard relevant information that we gain from our activity model. 0.18 0.16 Income $0-30k 0.14 Income $30-60k Probability Density 0.12 Income $60-100k Income $100k+ 0.1 0.08 0.06 0.04 0.02 0 $- $5 $10 $15 $20 $25 $30 Value of Time ($/Hour) SAN FRANCISCO COUNTY TRANSPORTATION AUTHORITY
- 38. Dynamic Traffic Assignment Assumptions Dynamic User Equilibrium: No vehicle can unilaterally shift paths and improve their generalized cost Generalized Cost: travel time + (left turn*left turn penalty) + (right turns*right turn penalty) turn penalties decrease round-about paths Additional Network Inputs: Signal timings Stop signs Intersection Geometry Saturation Flow Rates/Jam density SAN FRANCISCO COUNTY TRANSPORTATION AUTHORITY
- 39. Citywide DTA Network Calibration & Validation: Happening Now! SAN FRANCISCO COUNTY TRANSPORTATION AUTHORITY 39
- 40. That’s it! [email protected] www.sfcta.org/modeling SAN FRANCISCO COUNTY TRANSPORTATION AUTHORITY
Editor's Notes
- #8: Not shown:Initially Schedule Tours
- #9: Not shown:Tour scheduling based onAccessibility by time of day for chosen destination
- #23: We use a “Doubly Stochastic Route Search” to find other potential routes in the available choice setBovy, P. & Fiorenzo-Catalano, S. (2007), “Stochastic route choice set generation: behavioral and probabilistic foundations,” Transportmetrica 3, 173-189.
- #26: Talking points/or circle/or flip through
- #27: Yellow doesn’t stand out enough
- #28: What we’re doing now and why it’s not optimal (subjective which makes it hard to forecast, high-low, O/D-based)Goal: Represent effect of network projects and land use changes on pedestrian behavior. Better represent pedestrian level of service, the utility of the pedestrian path. Remove the subjectivity and use continuous variables.This presentation will show how we represent pedestrian utility for mode choice
- #29: Pretty consistent with previous findings (i.e. school purpose and indirectness)For transit, origin/dest refer to *tours*In general, the magnitude of all coefficients are bigger at the tour destination than the tour origin (including distance). Conditions of the walk at the tour destination matter more than at the tour origin (home).
- #31: These are based on Population Density and Indirectness (tour origin)- Talking point circles