Riding the 38-Geary - On a Computer

George Musser discusses Jun Yoshioka's computer model of SF's busiest bus line

One of RESCUE MUNI's main goals has been to make constructive suggestions for improving service without any additional money. Often, simple changes in the way a line is run can have dramatic effects on its performance.

To study these effects, RESCUE has been fortunate to work with Jun Yoshioka, who received his master's degree in operations research from San Francisco State University last spring. For his master's thesis, Yoshioka painstakingly programmed a computer simulation of the 38-Geary line, the busiest in the system. His model has allowed us to experiment with different schedules and operating procedures, so that we can see what might, and might not, help performance.

Yoshioka's computer divides the route into segments. The time it takes a bus to drive through a segment varies at random - it might be slower or faster depending on delays in passenger loading and traffic. Yoshioka based his model on Arena 2.1, a standard computer simulation program for Windows 95, and verified its accuracy by comparing its output to data from our riders' survey last February and his own timing measurements at Fillmore and Geary.

Yoshioka's initial goal was to examine how bus breakdowns and unscheduled driver absenteeism affect performance. To simulate breakdowns, the computer chooses a bus at random and holds it in one place for a specified time. To simulate absenteeism, the computer erases one of the buses before it can start service. Obviously, passengers wait longer when buses break down and drivers miss runs.

But the surprising thing is that the delays can be minimized by clever scheduling. Yoshioka compared Muni's present schedule with an alternative schedule. Unlike the current schedule, with its complex series of layovers, this alternative schedule simply requires a bus to leave exactly every 4 minutes from the Transbay Terminal and every 12 minutes from each of the three other terminals. The key is to hit the departure time exactly every 4 minutes.

This alternative, it turns out, greatly improves reliability. Buses are more evenly spaced, and most passengers can plan on waiting no more than 5 minutes even at stops well down the line. Even better, the alternative is more resistant to bus breakdowns and driver absenteeism. Compared with the present schedule, riders would experience fewer delays even when the same number of drivers failed to show up for work, or the same number of buses broke down. To top it off, the alternative schedule provides the same level of service with fewer buses.

It might sound too good to be true, but the basic principle is really quite simple: Be consistent. By sending a bus every 4 minutes from the end points, Yoshioka's alternative schedule puts buses to the most efficient use. To make the plan work in practice, Muni would probably need to park a backup bus and driver at Transbay and Ocean Beach. In case the regularly scheduled bus couldn't start its trip on time, the backup bus would leave instead. That way, a bus would leave every 4 minutes in all but the most unusual circumstances. Although the backup buses would be idle most of the time, the route overall would still require fewer buses.

Yoshioka and two other mathematically minded members, Andrew Sullivan and George Musser, met with Muni managers to discuss the model in June. The managers praised the model and said it could help them to pinpoint possible bottlenecks - locations where it might be good to put a street supervisor.

In a second meeting in August, Muni's computer guru asked Yoshioka to prepare a similar model for the 22-Fillmore line. Because the 22 line has only two end points, rather than the four of the 38, Yoshioka can redirect the limited capability of his computer into better simulation of bunching - the bane of the 22.

Hopeful though these meetings were, Muni has yet to follow up. It still prepares schedules using a '70s-vintage computer, which makes crude estimates of average running time, but does not perform any detailed simulation of actually driving a bus on city streets. Our eventual plan is to devise an improved schedule and convince Muni to adopt it.

Full details of this research can be found elsewhere on RESCUE MUNI's web site. We need volunteers to help with this project, either by going out on the street and timing buses or by helping Yoshioka hone his model. If you would like to work on this project, please leave a message for Yoshioka or Jeff Goldblat on RESCUE's hot line at 2731558 or transit1@ rescuemuni.org.

[ Transfer index | RM Home Page ]

RESCUE MUNI Transfer, Winter 1997-98.
Copyright © 1997 RESCUE MUNI. All rights reserved.
Questions? Send us