Validate GTFS-based speed estimates with realtime data

[dan-mccabe]

Overview
We use static GTFS and road network data to create estimated speed profiles for bus trips based on features like the scheduled time between stops, link distances, and speed limits. These might not align with real-world bus speeds. To validate and fine-tune our approach, we want to scrape some GTFS realtime data for a few transit agencies and compare our speed predictions to observed speeds.

Subtasks

• Write code to scrape realtime feeds (at reasonably high frequency) and save to NREL HPC system
• Write code to process archived realtime data to develop estimated real-world speed profiles for different trips. These trips should be matched to the OSM network for consistency with the routee-transit pipeline.
• Evaluate speed prediction accuracy (at OSM link level) of current routee-transit speed estimates. We want to have a fairly large sample of observations for each trip ID included to get a sense of uncertainty.
• Fine-tune our speed estimation pipeline to improve speed estimation performance.

Getting Started
As discussed in our RouteE-Transit team meeting, @robfitzgerald and I will coordinate on setting up the initial scraping process to gather realtime data. I propose we start by monitoring the following feeds:

• RTD, available at https://www.rtd-denver.com/files/gtfs-rt/VehiclePosition.pb
• King County Metro, available at https://s3.amazonaws.com/kcm-alerts-realtime-prod/vehiclepositions.pb

There isn't a firm standard for refresh rate on realtime vehicle positions, but 30 seconds is the absolute minimum. I'd suggest we try pulling the data every 10 seconds to start. Since we're trying to get good estimates of speed at the road link level, we want things to be pretty high-resolution.

I think it's okay to start by storing the complete vehicle position feeds at each time, and I will work to develop some code that can do some quick processing to filter out unnecessary data and only store what we need.

@robfitzgerald, please let me know what you think and if you need more clarification to help me get going on this.

[robfitzgerald]

set up a downstream issue in #21 to build tooling for GTFS realtime scraping