Whether you're heading in from San Francisco International Airport or out to downtown Berkeley, if you're taking rapid transit in the Bay Area then you're in for an analytics ride -- one that starts the moment you pass through the fare collection gate.
That's because on-time train service is the most important issue to Bay Area Rapid Transit's 370,000 daily riders, as one customer satisfaction survey after another shows, according to Roy Henrichs, manager of reliability engineering for the San Francisco Bay Area Rapid Transit District. And the transit agency can't possibly know if its trains are running on time and patrons arriving at their destinations as expected if it doesn't engage in some rather sophisticated operational analytics.
"You can do management by walking around -- but it doesn't work. You obviously need analytics," Henrichs shared with me during a recent phone interview.
His team conducts numerous types of operational analytics, including system performance analysis, passenger flow modeling (PFM), and delay analysis, as well as a variety of other modeling used for forecasting and what-not, he told me. And the data required for the analytics is complex and voluminous.
BART's data infrastructure comprises an IBM Maximo asset management system on an Oracle database/Linux platform. BART relies on code from SAS, this site's sponsor, to move the data around and do the analytics, Henrichs says, noting that the code was easily ported from the agency's previous mainframe-based environment .
With on-time service top priority, PFM is one of the most critical applications running in its environment. Using time series analysis, combined with econometric data, the team runs models aimed at forecasting ridership. The goal is making sure train schedules are optimized to keep customer satisfaction high while curbing BART's costs. "We don't want to run trains that are under- or overloaded… and the PFM, among other things, captures or estimates train loadings for use in generating the train schedules."
Henrichs's team can find out if a train that's supposed to arrive at Embarcadero Station at 10:17 a.m. actually showed up early, on time, or late. And because it also has data coming from its fare gates, it can monitor the motion of passengers through the system, too, using patron ID numbers. "What this department wants to know is whether Joe Citizen, who got on at Berkeley Station at 9:15 a.m. and off at Embarcadero Station at 10:45 a.m., arrived on time or not... This is not a Big Brotherish kind of thing."
Rather, he said, it allows BART to address the three stages a passenger goes through when arriving at a station. "You arrive at the station, and the first question you ask is, 'Where's my train?' Then you ask, 'Where's my seat?' And finally, 'Will I be on time?' "
With its operational analytics, BART is able to get to a "Yes" answer more often than not and keep satisfaction high.