Fort Lauderdale, Florida, United States
Open Source Software in Public Transportation: A Case Study
Summary Information
Public transportation agencies increasingly use software to support mobility services. This may include trip planners, geocoders, and timetables. A report at the 2020 TRB conference by analysts at the Center for Urban Transportation Research performed a study looking at how open-source software (OSS) has evolved into production deployments at transit agencies, including a look at the risks and opportunities when compared to closed-source software. To research this area, the analysts interviewed over a dozen public and private sector stakeholders, reviewed Federal Transit Administration's Mobility on Demand (MOD) Sandbox projects, and investigated perceived risks of OSS. They developed a list of recommendations based on their research.
Lessons Learned
The travel time based (TT-based) algorithm was found to be the most effective of the dynamic tolling schemes.
- In the calibrated logit model, it performed slightly, though not significantly, better than the density-based algorithm, and significantly better than the travel time savings based (TTS-based) algorithm.
- In the more sensitive logit model, it performed significantly better than both other algorithms.
- In both scenarios, it consistently ranked as having the most effective capacity utilization across all scenarios, and was found to have the lowest total network delay the majority of the time. The authors of the presentation noted that the TT-based algorithm was actually the least sensitive to input, increasing tolls less frequently, but attracting more vehicles to the Els than the other two algorithms.
- The TTS-based algorithm was found to be the most effective at meeting the policy standard, which called for maintaining a speed of at least 45 mph among 95 percent of all vehicles in the EL. While the TTS-based algorithm met this criterion every time, the TT- and density-based algorithms failed to meet it 25 percent of the time.
- The TTS-based algorithm was also found to perform better at the 15-minute evaluation mark than at the 5-minute evaluation mark, as its response was slightly slower than the other two algorithms.
