By Rahul Nair
Miami Dade Transit (MDT) serves the seventh largest urban population in the U.S. As you might expect, public transportation for 2.7 million people can be challenging, but especially when asked to operate at a budget deficit that’s projected to reach $81.4M by 2023.
For MDT, operating as efficiently as possible is an imperative. That’s why in 2014 the organization turned to IBM to help identify and eliminate costly inefficiencies. In particular, it asked my team and me to help solve a growing dilemma on the streets of Miami’s South Beach: bus bunching.
Bus bunching occurs when two or more busses serving the same route arrive at a stop at the same time. Such phenomena results in longer wait times for passengers and dramatically lowers the utilization of the busses, costing cities millions of dollars a year.
To figure out how to get control of the situation, we launched an eight-week pilot program aimed at reducing delays and improving services. The pilot targeted four main routes along the South Beach corridor and provided real-time predictive analytics of system performance.
Our prediction system provided alerts of potential bunching events to the operations control up to an hour in advance, expanding their ability to anticipate, manage, and ultimately prevent bunching. A key secondary use of such operational data provided data insights that led to better planning, scheduling, and performance management.
For large transit systems, the volume of real-time data can grow quickly, on the order of hundreds or thousands of position updates per second. Key measures for network-wide awareness that require complex analytics have large real-time computational needs. In contrast to “data-at-rest” applications, where data is stored, loaded in memory, and then processed, a streaming data model processes data as it is received.
IBM Infosphere Streams middleware platform allows for parallel and high performance stream processing that scales over a range of hardware environments. Based on computational requirements, applications are distributed over any sized server cluster, and analytics are executed in memory for low latencies and high data throughput.
The pilot used our Intelligent Transit Analytics based on IBM Infosphere Streams. It analyzed more than 1.4 million bus positions, representing 19,974 bus runs between December 4, 2013 and March 18, 2014. The operational data allowed a rich characterization of service characteristics, its variability, and pinpointed bottlenecks within the system.
As the IBM Intelligent Transit Analytics solution received bus position updates, it performed rigorous quality checks including on-route/off-route analysis through map-matching, and de-noising of the position data to ensure consistency. It then recorded how far a bus was from the start terminal and the time it reported its position. A collection of the bus’s positions creates a trajectory – a path in space and time defined by a bus as it periodically reports its position.
The distance from a start terminal, also called offset, effectively measures bus position as if the route were laid out as a straight line. A plot of a bus’s trajectories, known as a space-time diagram, was generated to represent the results and assist in a visual manner to examine bus performance. Each line of this space-time diagram shows an individual bus’s journey alongside all bus journeys observed for the day. The slope of the trajectory denotes the speed – with steeper lines indicating faster bus speeds. When bus trajectories meet on the diagram, it indicates bus bunching locations and times, along with mid-day slowdowns for several buses.
Buses were prematurely returning to their scheduled routes which originated in the downtown terminal. This meant that drivers were only able to stop at the terminal for a few minutes. Consequently, their journey times were doubled as they were not able to stop for a break. Due to this shortened stoppage time, a 35 km route with 113 stops becomes 70 km with 226 stops. Delays from 6 to 11 minutes on these bus routes ultimately cost the city more than $4 million in operational costs.
At the end of the pilot, our report to MDT provided key performance indicators of bus bunching, assessed accuracy of prediction of bus arrivals and bunching event alerts, and provided a detailed assessment of MDT service performance and produced schedule refinements. The operational real-time recommendations were verified and used by the MDT operations team – and they are now armed with predictive capabilities to help keep the busses on schedule.