Modifying the Seed Matrix in the Iterative Proportional Fitting Method of Transit Survey Expansion

Transcription

1 Modifying the Seed Matrix in the Iterative Proportional Fitting Method of Transit Survey Expansion Sujith Rapolu Ashutosh Kumar David Schmitt Innovations in Travel Modeling (Baltimore, MD) April 30, 2014

2 Outline Introduction Case Study Conclusions April 30, 2014 Page 2

3 Transit Surveys Transit surveys - Travel patterns and characteristics of existing riders - Critical role in calibration and validation of models April 30, 2014 Page 3

4 Transit Survey Expansion Traditional expansion to total ridership by direction, time of day and route. - Accuracy of O-D flow estimates?? Iterative Proportional Fitting (IPF) method to expand transit surveys - Seed matrix is scaled iteratively to match row and column totals - What about cells with no observed records?? Addressing missing interchanges in the seed matrix of IPF method April 30, 2014 Page 4

5 CASE STUDY April 30, 2014 Page 5

6 Tri-Rail Overview 72-mile commuter rail system in Southeast Florida 17 stations across 3 counties 1hr 45min traveling time 50 trains/day service 14,300 average daily riders (2013) Not easily accessible to any major attraction by walking April 30, 2014 Page 6

7 Survey Data Survey conducted in February trains surveyed (14 in each direction) Station level door counts on all trains on the survey date 3,739 useable records Final sample rate of the cleaned survey - 26% Final IPF expansion by time of day and direction - Six seed matrices (3 time periods x 2 directions) - IPF accomplished using FRATAR program April 30, 2014 Page 7

8 Seed Matrices for IPF Station-to-station seed matrices are obtained from survey records Control totals are station boardings and alightings Some station-to-station interchanges had zero surveyed trips Total no. of cells with zero surveyed trips: One-station interchanges account for 60 cells Quick fix: Replace cells in the seed matrix having no observed trips with one synthetic record April 30, 2014 Page 8

9 Result (1-station trips) April 30, 2014 Page 9 Replacing zero cells with 1 Replacing zero cells with 0.5 Records observed from survey Resulting from IPF Total Cells with non-zero observed trips Cells with zero observed trips % of 1-station trips from cells with zero observed trips 51% 42% Total ridership 14,300 14,300 % of 1-station trips 6% 5% Past surveys showed that 1-station trips account for only 2-3% of total ridership Replacement of zero cells with 1 s over-estimates by 100%

10 Modifications to Seed Matrices 1-station interchanges with no observed trips left unchanged Every other interchange (2+ station movements) with no observed trips given a value of synthetic records were added April 30, 2014 Page 10

11 Final IPF Expansion 1-station trips Records observed from survey Replacing zero cells with one-station trips (2.6% of the total) Replacing zero cells with 0.5 3,739 useable records got expanded to 13,770 trips Final Expansion Resulting from IPF Total Cells with non-zero observed trips Cells with zero observed trips % of 1-station trips from cells with zero observed trips 51% 42% 0% 89 synthetic records got expanded to 530 trips (3.7% of the total) April 30, 2014 Page 11

12 Trip Characteristics of Synthetic Records EXAMPLE PM Southbound Mangonia Park West Palm Beach Lake Worth Boynton Beach Delray Beach Boca Raton Used the 51 records to generate the trip characteristics April 30, 2014 Page 12

13 Trip Characteristics of Synthetic Records (Contd.) EXAMPLE Trip purpose, access mode and market segment are the trip characteristics assigned One synthetic record between Boynton and Boca Raton is split: weight for HBW purpose, walk access, zero-car household weight for NHB purpose, PNR access, 1 car household weight for HBNW purpose, KNR access, zero-car household... April 30, 2014 Page 13

14 Result (IPF vs. Traditional Expansion) Traditional expansion to station boardings (ONS) by direction and by period (ONS+OFFS)/2 Difference with observed Station name Observed IPF Traditional IPF Traditional Mangonia Park 1,056 1,045 1, West Palm Beach 1,208 1,195 1, Lake Worth Boca Raton 1,488 1,464 1, Pompano Beach Cypress Creek 1,030 1, Fort Lauderdale Airport Golden Glades Opa-locka Metrorail Transfer Station 1,238 1,198 1, Hialeah Market/MIA Only stations with a difference of at least 5% are shown April 30, 2014 Page 14

15 Conclusions IPF expansion provides a better match to observed station activity IPF expansion accounts for O-D flow movements ignored in the traditional method of expansion Important to properly address the missing movements in IPF expansion Assign trip characteristics to synthetic records Future Research Determine the weight of all synthetic records being added to the seed matrix April 30, 2014 Page 15

16 Acknowledgements South Florida Regional Transportation Authority (SFRTA) Florida Department of Transportation (FDOT) April 30, 2014 Page 16

17 THANK YOU! April 30, 2014 Page 17