Amtrak Timekeeping Analysis

[Ryan]

I mentioned this in another thread, but someone asked about the Sunset Limited into NOL, so I figured that it was worth a new thread.

Over the last few days I've been playing around with learning Python. As a project, I've set after taking the text files that Amtrak publishes with the train status for a day (one displayed below), parsing the times into a useful format and then storing the data such that I can manipulate, analyze and plot it.

It's been a cool learning project, and it wouldn't have been possible without John Bobinyec's Amtrak Maps site, which archives and makes available all of the text files.

There's an Amtrak Delay's site, but you can only look back 4 weeks at a time - once I'm done, I'll have the status of a train from 2006-2012 (and later, as John makes the text files available).

Here's what the Sunset Limited's arrival times into New Orleans look like. I shifted the data series when the schedule change took effect. It didn't do much to timekeeping until the early September timeframe, when something happened to make it significantly more better.

In addition to graphs, I can also pull stats out - for example in the other thread, I was able to show that the Palmetto arrived too late for the guy to make his bus connection less than 1% of the time.

I haven't worked it through yet, but I'd also like to put together plots where delays are plotted along the stations on a route on the x axis - by plotting more than one day's train at a time, eventually I'll end up with a spaghetti plot that may show some interesting trends. I'm thinking the first train I'm going to look at is the Cardinal and see just how bad the BBRR looks.

If there are other trains and connection's people are interested in, I'm happy to take a look. This isn't probably something that I'll ever have polished enough to turn into something for public use (I'm sure that anyone that has any formal training in programming would look at my mess of code and recoil in horror), and each step of the way I'm increasing the automation and error handling to make things go a little faster (I probably spent about 2 hours transitioning from looking at the Palmetto to the Sunset Limited).

So, what else do you think would be interesting to look at?

 

[Nathanael]

How often does a train from the west (any of them) arrive late enough that the Lake Shore Limited departure will be delayed for it. Yeah, I know, that's extra work, you have to find the worst delay of five trains...

 

[TimSpencer]

Amtrak timekeeping.... hmm.....

Do those two words really belong together in a sentence?

I always thought Mr. Cosmo Kramer of Kramerica Industries

was in charge of timekeeping at Amtrak.

 

[Anderson]

Missed once-daily connections would be interesting. I'd really like to see Acela-to-Adirondack connections missed NB and/or Adirondack-to-9:15 Acela connections made (I'd call it the 2115 Acela, but I am reasonably certain there's an actual Acela bearing that number).

Edit: I'd also be curious to see Cap-to-Star connectivity, too, now that I think about it.

Edit 2: And what made the Palmetto 15 hours late one time?

 

[Paulus]

Edit 2: And what made the Palmetto 15 hours late one time?

Probably a derail and/or trackwork in a bad area for it. I've seen a derailed freight result in the Sunset Ltd being 36 hours delayed into Los Angeles. Though apparently the Palmetto managed to be 22 hours late into Charleston back in 2004.

Meanwhile, CSX reported signal and switch failures were playing havoc with Atlantic Coast schedules. Amtrak’s southbound Palmetto, train No. 89 of January 27, was seriously delayed in the 70 miles between Dillon, Florence and Kingstree, S.C., after which the crew “blew up” after completing 12 hours on-the-job. Their hours of service expired just south of Kingstree, and a relief crew had to be sent to the stopped train. The Palmetto arrived in Charleston some 22 hours late.

 

[afigg]

Here's what the Sunset Limited's arrival times into New Orleans look like. I shifted the data series when the schedule change took effect. It didn't do much to timekeeping until the early September timeframe, when something happened to make it significantly more better....

I haven't worked it through yet, but I'd also like to put together plots where delays are plotted along the stations on a route on the x axis - by plotting more than one day's train at a time, eventually I'll end up with a spaghetti plot that may show some interesting trends. I'm thinking the first train I'm going to look at is the Cardinal and see just how bad the BBRR looks.

It might be useful to look at each LD train into Chicago over a year or even 2 year period. Likely to see seasonal variations in OTP. Spring rains or floods for the western LD trains, winter storms, track work season, etc. The Sunset Limited into NOL might have gotten better due to the completion of track work projects.
To see the effects of the BBRR on the Cardinal, you could plot the OT numbers for the eastbound Cardinal a 2 lines: Clifton Forge and Culpepper or Manassas. I've noticed that a late Cardinal will sometimes lose more time between Culpepper and Manassas. presumably because it is out of the slot and gets stuck in the single track segment. But an overall chart of the Cardinal over the past year or longer into WAS (where it usually makes up time to NYP) and CHI will show better OTP since last fall.

Something else that could be graphically shown is the padding at the end of an LD train route. Plot the OTP at the last station which picks up passengers prior to the final stop and another line for the final stop.

Another one for fun would be the Pennsylvanian into Pittsburgh over the past year. How often does it not get in 15 or 20 minutes early?

 

[JoeBas]

I was the SL to NOL requestor, was just wondering what my odds of actually making use of the Hilton connecting between SL and Crescent the next morning were since the schedule change. Thanks!

 

[Ryan]

Looks pretty good - I don't know what changed in September (or how that's held up into the new year), but at least in the SEP-DEC timeframe it looks pretty sweet.

Thanks for the rest of the ideas. John suggested to me that the Amtrak provided "3 Hours Early" text at the end of each line is potentially more accurate. I'm going to work towards implementing that before I do any more analysis. I think I've worked out in my head how to parse "1 hour and 30 minutes late." into "90" for plotting.

 

[Ispolkom]

A connection I'm personally interested in is #14's arrival at PDX, since there is barely an hour between its scheduled arrive and #28's departure.

 

[Nathanael]

It is proposed (in the Performance Improvement Plan) that the eastbound LSL out of Chicago shift to a 6:00 PM departure I'd therefore also be curious how often each of the western trains (and the CONO) arrive after, say, 5:50 PM -- in other words, how often they are so late that it would create an issue for passengers transferring to the early-departing LSL.