A Real-Time Service Management Decision Support System for Train Dispatching at New York City Transit

Abstract

With ridership near modern highs, New York City Transit's (NYCT) subway network frequently operates at or near capacity. This makes maintaining a high-quality service both challenging, due to the lack of ''slack,'' and exceptionally important, due to the large number of riders affected by disruptions. To this end, train dispatchers constantly monitor the network and adjust service to respond to delays. This paper presents a decision support system developed by NYCT which uses real-time train movements and historical ridership information to provide dispatchers with recommendations for holds and station skips in real time. The system uses heuristic headway criteria to determine hold or skip candidate trains, and then estimates the net passenger time savings of each potential hold or skip using estimated origin-destination flows and basic assumptions about passenger behavior. Potential actions that meet a passenger benefit threshold are recommended, and communicated to dispatchers with a simple dashboard. A pilot implementation of the system has been in use at NYCT's Rail Control Center (RCC) for several months, though many details of the system are still in development. Initial observations indicate the system is helping dispatchers manage train service more effectively, producing large passenger time savings.

Algorithm for Tracing Train Delays to Incident Causes

“…Though this information is not strictly needed for each stop to determine a new incident, as will be seen below, it is useful for related analyses of why a new incident could not be assigned. As shown on the first row of Figure 2, they were defined as: Delay : The typical running and dwell times are subtracted from the actual times to get the delay (or time savings, if the trip is faster than typical) at each station. Adjacent Train Impact : True if the train falls into any of the following categories related to whether the adjacent train’s movements could have caused a delay: Delay in that segment involved running time and the leader train was less than 3 min ahead Delay in that segment involved dwell time and either: The leader train was less than 3 min ahead There is a potential hold because of the leader, based on the logic of NYCT’s service management tool, SIRE, and the leader is less than 4 min ahead ( 12 ) There is a potential hold because of the follower Both trains had dwell times of at least 5 min at the same time Adjacent Train Incident or Non-Revenue : True if the adjacent train has been assigned to a root-cause incident (that has already occurred) or if the adjacent train is a non-revenue train (work train, layup train etc.) Then the train is analyzed to narrow down potential causes of delay.…”

“…Adjacent Train Impact : True if the train falls into any of the following categories related to whether the adjacent train’s movements could have caused a delay: Delay in that segment involved running time and the leader train was less than 3 min ahead Delay in that segment involved dwell time and either: The leader train was less than 3 min ahead There is a potential hold because of the leader, based on the logic of NYCT’s service management tool, SIRE, and the leader is less than 4 min ahead ( 12 ) There is a potential hold because of the follower Both trains had dwell times of at least 5 min at the same time …”

“…Delay in that segment involved dwell time and either: The leader train was less than 3 min ahead There is a potential hold because of the leader, based on the logic of NYCT’s service management tool, SIRE, and the leader is less than 4 min ahead ( 12 ) There is a potential hold because of the follower Both trains had dwell times of at least 5 min at the same time …”

“…There is a potential hold because of the leader, based on the logic of NYCT’s service management tool, SIRE, and the leader is less than 4 min ahead ( 12 )…”

Automated Train Identification and Train Position Monitoring at New York City Transit

“…Dispatchers who previously operated remotely in local master towers with information only about anonymous track occupancy within the interlocking controlled by their tower could now see exactly which trains occupy track circuits from terminal to terminal, and routes are automated based on the schedule with the potential to manually override them in the electronic system for service management. The system, known as Automatic Train Supervision (ATS-A), enabled a more holistic response to service management along with many ancillary benefits, including countdown clocks, more accurate reporting of performance metrics ( 5 , 6 ), and development of internal service management tools ( 7 , 8 ).…”