ESTIMATION OF PRE-COVID19 DAILY RIDERSHIP PATTERNS FROM PAPER AND ELECTRONIC TICKET SALES DATA WITH ORIGIN-DESTINATION, TIME-OF-DAY, AND TRAIN-START DETAIL ON A COMMUTER RAILROAD: QUICK-RESPONSE BIG DATA ANALYTICS IN A WORLD STEEPED WITH TRADITION

摘要

Our niche method independently estimates hourly commuter rail station-to-station origin-destination (OD) matrix data each day from ticket sales and activation data from four sales channels (paper/mobile tickets, mail order, and onboard sales) by extending well-established transportation modelling methodologies. This algorithm s features include: (1) handles multi-pack pay-per-ride fare instruments not requiring electronic validation, like ten-trip paper tickets "punched" onboard by railroad conductors; (2) correctly infers directionality for direction-agnostic ticket-types; (3) estimates unlimited ride ticket utilization patterns sufficiently precisely to inform vehicle assignment/scheduling; (4) provides integer outputs without allowing rounding to affect control totals nor introduce artifacts; (5) deals gracefully with cliff-edge changes in demand, like the COVID19 related lockdown; and (6) allocates hourly traffic to each train-start based on passenger choice. Our core idea is that the time of ticket usage is ultimately a function of the time of sale and ticket type, and mutual transformation is made via probability density functions ("patterns ") given sufficient distribution data. We generated pre-COVID daily OD matrices and will eventually extend this work to post-COVID inputs. Results were provided to operations planners using visual and tabular interfaces. These matrices represent data never previously available by any method; prior OD surveys required 100,000 respondents, and even then could neither provide daily nor hourly levels of detail, and could not monitor special event ridership nor specific seasonal travel such as summer Friday afternoons.