Great Western Coffee Shop

At Chippenham, yesterday evening:

(http://www.wellho.net/pix/imposs.jpg)

Information systems can't always be certain about delays, especially during ongoing incidents ... but when they suggest something that simply couldn't / wouldn't happen, they make it much harder to trust them.

In the example above, you have a 125, express from Swindon, leaving Chippenham at 19:14 and then a slower train (a 150 or 158) that's followed it from Swindon leaving 2 minutes later.  Of course, what we were being told changed .. the Bristol train arrives at about 19:18, and the Southampton train called at around 19:25.  Now - I could have guessed that's what would happen at 18:56 - shouldn't the computerised information systems, with the skill of industry experts, be able to do as well (or better) than my amateur predictions from just a bit of knowledge of the line layout and trains in use?

Indeed, I have raised this previously with train company staff. We passengers are encouraged to believe that the "Expected" column on the departure screens is a forecast. It isn't a forecast at all, it merely reflects the punctuality of the train at the last time it passed an automated reporting point (except in rare cases where manual intervention takes place). And I'm sure many of us have seen instances where a train has "disappeared" from the departure board before it's even turned up, because a delay between reporting points has not been captured and the system assumes that the train is happily on its way, although this seems to have improved and trains are often simply shown as "delayed".

To provide a forecast, I guess the live departures system would need to have some way of modelling forward the  likely timing of the train based on the operational conditions, the positioning of other trains and signalling policy for allocating trains to slots (and probably lots of other variables). It may not be as straightforward to do as it seems. Fortunately the current system works fine in the majority of situations, so it may have to be something we have to live with for the moment.

Quite apart from the fact that the 1855, showing on time, is already nearly 2 minutes late.

I would go as far to say that it would be impossible to do perfectly.  There are far too many variables.  A few tweaks here and there could improve things a little though.

The information systems are rather optimistic sometimes about how much of the delay trains will make up. However I'd say it's better to predict the train arriving earlier than it does than predicting it will arrive later than it actually does, in which case people could have left the platform and potentially miss the train.

And you get instances when a starting train is delayed and nobody manually inputs an estimated departure time resulting in the train disappearing from the screen after a few minutes.

Tricky one this: I understand exactly what grahame means, but I think the practicalities would rule it out. I tend to view those "expected" arrival times as exactly what they are, which is to say the scheduled departure time of the train with its current delay factored in, rather than a prediction of what time they may actually arrive.

I suspect that there are so many variables that it wouldn't be possible: it's difficult to predict just how long a train that's out of its booked path might get delayed, depending on what it's following, for example.

Southern Stag makes a very good point: I would say that it's generally better to give a slightly optimistic delay forecast so that people don't go wandering off then return to find the train has made up time and they missed it.


Different TOCs specify different thresholds: departure boards at SWT stations will show delays of 1 minute and above. However from observation, FGW-managed departure boards will show a train as "on time" until it's delayed for 3 minutes or more.