Selecting cars for loading + statistical observation


devansprr
 

--- In STMFC@..., Malcolm Laughlin <mlaughlinnyc@...> wrote:

I took a couple of OR courses when I was at MIT. What I found in
the real world of railroads was

1) The distributions were so lumpy that statistical methods
usually didn't work.
Malcom,

Missed the significance of this on my first read, but to me this would
help demonstrate that large variations from the national average, on
any given train, on any given day, and in any given yard, could be huge.

This gets back to my initial concern - we may need to make sure our
trains don't look "too" average. For rolling stock switched during an
op session, let things fall where they may, and perhaps stay near the
national average for deliveries from staging, but for a run-through
consist, or a large block that transits a yard but is not spotted to a
customer(s), but instead runs back into staging, hitting the average
all the time may not look prototpical.

Thoughts?

Thanks,
Dave Evans


devansprr
 

--- In STMFC@..., Malcolm Laughlin <mlaughlinnyc@...> wrote:
I can't tell you some amusing tales about trying to apply linear
programming to car distribution because that clearly required after
1961 computer capability.

Malcom,

Thanks for the insight from the OR perspective. Did anyone ever try to
use OR to analyze historical data to see if they could spot some
unknown operational problem/trend rather than solve an immediate
problem? I would expect not since the people on the ground probably
knew what the problems were - very different from telephones where the
lack of human involvement (relative to the number of "events") led to
a lack of information.

Appreciate the programming humor. As an engineer that graduated about
the time the PC came out, I'm continually amazed at how much analysis
was accomplished during the days before computers. Lots was done in
the 60's and 70's with main-frames and mini-computers, but before that
- Wow! and often with very large data sets.

Thanks for the help,
Dave Evans


Malcolm Laughlin <mlaughlinnyc@...>
 

My apologies for taking so long to replay to all of these interesting posts. On top of an extemely busy week, the hard disk on my office computer crashed and I'm just now catching up.

I’d like to address Dave Evans’ comments about how cars were loaded.


1) and the other for Harrisburg, and he has a bunch of empty 40 foot XM's - including one PRR, and one NYC, would he just flip a coin? Or do the agents not know the load's destination, only that a shipper needs a 40' XM?
-----------

In theory the agent should ascertain from the shipper where the loads were going and select cars accordingly. Among the reasons that the desired result was not achieved.


It would have been necessary to reswitch the empty box car track to get the right cars to the right shipper. Not likely to happen. The first ten cars go to shipper A, the next five to B, etc.
No agent was going to refuse bills of lading because the shipper put the Elkhart load in the PRR car and the Altoona laod in the NYC car.
Looking at it from the point of view of the shipper’s loading dock. Say there are ten cars going to ten consignees with similar loads. The loading dock foreman has ten loading orders. He puts the first order in the first car and the tenth order ends up in the 10th car. So the car that gets each laod depends on the sequence in which the orders were typed in the sales office. Nothing to do with marks of cars.


2), is it more likely that the NYC yard clerks would have it routed over the NYC's water-level route, rather than to the PRR, which could get it there over a slightly shorter distance
Yard clerks did not route cars. The freight agent receiving the shipper’s bill of lading copied the routing on the bill. If the shipper specified railroads but not junctions, the agent chose the junction giving his road the long haul, provided it was a service route. If the shipment was completely unrouted by the shipper, he would give his road the long haul and select the rest of the route based on service and friendly connections.

> 3) There is still the issue of statistical variations……….A little research found that Bell Labs/AT&T spent a ton of money developing the operations research for modeling these sorts of problems (it directly impacts the size of telephone systems). Turns out the deviations from average are often significant for many processes, especially if they involve processes that are not purely random

I took a couple of OR courses when I was at MIT. What I found in the real world of railroads was

1) The distributions were so lumpy that statistical methods usually didn’t work.

2) You couldn’t get any meaningful data from the field to plug in to the models.

3) What data you could get would be far enough behind that the results were out of date before transmission to the filed.

For example, I was asked to work on a model for diverting MDT reefers from the NYC main line to Weehawken for banana loading. But the railroad had a daily cycle imposed on a weekly cycle. Maybe there was a boat arriving every eight days. But the standard deviation of time between arrivals was two days and might be affected by labor in Central America and weather at sea. Now try to develop an algorithm for diverting the cars to Weehawken. Oh, and other variability – the flow of cars past the diversion point (Selkirk, congestion state of the yard and variations in train operations.


I actually did get some useful results with graphical inventory control models, but nothing more sophisticated was worth doing.

I can’t tell you some amusing tales about trying to apply linear programming to car distribution because that clearly required after 1961 computer capability.



Malcolm Laughlin, Editor 617-489-4383
New England Rail Shipper Directories
19 Holden Road, Belmont, MA 02478

[Non-text portions of this message have been removed]