Can computer modelling be used in the testing and designing of timetables on a large scale?

[HSTEd]

Mod Note: Posts #1 - #55 originally in this thread.

Has he got a box of magic tricks?

IIRC he wasn't in post for the 2018 timetable change fiasco, so he could do with taking a look at what happens when people tried to implement changes in shorter than the normal timescales.

I don't disagree that it must be frustrating with the time it takes to get changes implemented, just wonder what (if anything) he proposes to do differently?

Well one can imagine, with the computing power at our disposal, we could build a model of the entire railway network and then automate timetable generation based on optimisation for various outputs that would be put into the system.
Would take quite a while to build the physical model and get it set up, but I imagine you could generate timetables in a matter of days, given a fast enough computer.

 

[ForTheLoveOf]

Well one can imagine, with the computing power at our disposal, we could build a model of the entire railway network and then automate timetable generation based on optimisation for various outputs that would be put into the system.
Would take quite a while to build the physical model and get it set up, but I imagine you could generate timetables in a matter of days, given a fast enough computer.

It's not a problem I can see today's computer's or technology being capable of solving. There are simply far too many inputs and outputs, whose relationship is very difficult to formally put down into 'computer speak'. There are tricks that only people who have worked with the timetable for many years know. And for one, even with a resurrected 'BRB'-style system, you would presumably still have FOCs that have completely different interests to the 'BRB', which the algorithm would have to solve.

 

[306024]

It's not a problem I can see today's computer's or technology being capable of solving. There are simply far too many inputs and outputs, whose relationship is very difficult to formally put down into 'computer speak'. There are tricks that only people who have worked with the timetable for many years know. And for one, even with a resurrected 'BRB'-style system, you would presumably still have FOCs that have completely different interests to the 'BRB', which the algorithm would have to solve.

A computer will only generate an answer according to the rules input to it. If it was that easy Network Rail’s Timetable Planning System would be more advanced than it is. It is not a quick win by any stretch and way beyond the tenure of any minister.

And a timetable is so much more than just a bunch of times. It is platform working, parallel moves over junctions, unit diagrams, rolling stock stock maintenance, traincrew diagrams, traincrew rosters, performance regimes, the list could go on.......and on........

Changing franchises to concessions will not change the timetabling process overnight, reducing the numbers of players on the pitch would help, but will that ever happen?

 

[HSTEd]

It's not a problem I can see today's computer's or technology being capable of solving. There are simply far too many inputs and outputs, whose relationship is very difficult to formally put down into 'computer speak'. There are tricks that only people who have worked with the timetable for many years know. And for one, even with a resurrected 'BRB'-style system, you would presumably still have FOCs that have completely different interests to the 'BRB', which the algorithm would have to solve.

Well with enough computing power you could try and Montecarlo it.

You can urn tens or hundreds of thousands of simulations in parallel.

 

[306024]

Well with enough computing power you could try and Montecarlo it.

You can urn tens or hundreds of thousands of simulations in parallel.

Yes indeed. But what would be the decision criteria for determining which solution is the best result, and for whom?

 

[HSTEd]

Yes indeed. But what would be the decision criteria for determining which solution is the best result, and for whom?

Well you would pull out all the simulations that meet your decision criteria, and fit within the current requirements for rolling stock/staffing etc.
Chosing between fifty or a hundred cases of workable timetables is considerably less troublesome than generating one from scratch!

You could also apply a metric for "closeness" of train times to the previous one.

 

[markymark2000]

And a timetable is so much more than just a bunch of times. It is platform working, parallel moves over junctions, unit diagrams, rolling stock stock maintenance, traincrew diagrams, traincrew rosters, performance regimes, the list could go on.......and on........

Traincrew diagrams and rosters are a TOCs problem surely. Nothing to do with NR in the actual timetabling process. Rolling stock maintenance is also a TOCs issue and shouldn't be anything to do with NR.

 

[Domh245]

Traincrew diagrams and rosters are a TOCs problem surely. Nothing to do with NR in the actual timetabling process. Rolling stock maintenance is also a TOCs issue and shouldn't be anything to do with NR.

Thinking like that is how you end up with Thameslink ~May 2018. All well and good for NR to go "we've done our bit" but if the TOCs can't make it work then the end result is still unhappy passengers and cancelled trains

 

[ForTheLoveOf]

Traincrew diagrams and rosters are a TOCs problem surely. Nothing to do with NR in the actual timetabling process. Rolling stock maintenance is also a TOCs issue and shouldn't be anything to do with NR.

That's precisely the kind of compartmentalised thinking that represents all that's wrong with the industry at the moment. The aim isn't "how can we run the best possible railway given £Xbn", it's "how do we make our life easier". Allegedly this new 'Fat Controller' set-up will avoid this, but until there are details it's difficult to be convinced.

 

[markymark2000]

Thinking like that is how you end up with Thameslink ~May 2018. All well and good for NR to go "we've done our bit" but if the TOCs can't make it work then the end result is still unhappy passengers and cancelled trains

That's precisely the kind of compartmentalised thinking that represents all that's wrong with the industry at the moment. The aim isn't "how can we run the best possible railway given £Xbn", it's "how do we make our life easier". Allegedly this new 'Fat Controller' set-up will avoid this, but until there are details it's difficult to be convinced.

I can see the benefits of it all being included but it will make the process a lot harder and longer to sort. You need a main timetable creating and then staff worked around that (with changes made if they have to be to accommodate staff and unit diagrams). Diagrams and the timetable have to be planned at around the same time but you do also need to keep some things separate to avoid NR doing all the work. TOCs can schedule things in a certain way to make things as efficient as possible. NR can and does plan some thing ineffectively and therefore costs more money in the end.
The more you give to one company, the more there is which could fail in one go.


The Thameslink stuff was due to a huge service change as well as issues with staff. Everyone has a part to play and NR can't just say 'we've done our bit, work with it', they do need to be able to adapt if train crew doesn't work but planning so much together means it can be less efficient and a railway costing more money isn't very good.

 

[ainsworth74]

It's almost like trying to completely divorce steel rail from steel wheel wasn't actually a terribly good idea...

 

[mmh]

I can see the benefits of it all being included but it will make the process a lot harder and longer to sort. You need a main timetable creating and then staff worked around that (with changes made if they have to be to accommodate staff and unit diagrams).

I'm usually reluctant to defend the "but we don't do it like that" take of some staff, but it's pretty obvious you can't design a sensible timetable without thinking of staffing. With long routes and many remote outposts the staff still need to get back to their base and you want to make it reasonably efficient in staff numbers.

 

[ForTheLoveOf]

It's almost like trying to completely divorce steel rail from steel wheel wasn't actually a terribly good idea...

There's so many accounts of the privatisation process that each come to roughly the same conclusion. Privatising BR wasn't necessarily a terrible idea per se. But the dogmatic approach to the way it was done has tarred the railways ever since, and in some cases led to lives being lost (let alone livelihoods).

The splitting up of freight into lots of different chunks "to encourage competition", only for EWS to end up as the sole bidder (with Freightliner being a buyout), says it all really. Admittedly we now have several different major FOCs but as they only pay variable track access charges, which don't even come close to covering a fair share of NR's costs, they are all in effect subsidised by the Government. 'Twas ever thus since privatisation...

 

[Mag_seven]

It's almost like trying to completely divorce steel rail from steel wheel wasn't actually a terribly good idea...

Indeed as most people in the industry at the time would have told you that. There was some stuff about an EU directive about separating infrastructure from operations as a justification but that was only required in an accounting sense.

 

[HSTEd]

I can see the benefits of it all being included but it will make the process a lot harder and longer to sort. You need a main timetable creating and then staff worked around that (with changes made if they have to be to accommodate staff and unit diagrams). Diagrams and the timetable have to be planned at around the same time but you do also need to keep some things separate to avoid NR doing all the work. TOCs can schedule things in a certain way to make things as efficient as possible.

A timetable that can't be provided within established staff and rolling stock strength is pretty useless though!
People want their to be a separation there but there isn't really one.

A railway is a cohesive whole, you can't just split it into arbitrary chunks and expect it work properly.

Indeed as most people in the industry at the time would have told you that. There was some stuff about an EU directive about separating infrastructure from operations as a justification but that was only required in an accounting sense.

An accounting sense alone does untold damage.
It makes it considerably harder to coordinate properly, indeed coordination under that system is specifically forbidden to support open access operations.

 

[306024]

Traincrew diagrams and rosters are a TOCs problem surely. Nothing to do with NR in the actual timetabling process. Rolling stock maintenance is also a TOCs issue and shouldn't be anything to do with NR.

As a couple of others have explained, crew diagrams are integral to timetable construction. To obtain the most efficient timetable, planning isn’t simply a linear process of timetable -> unit diagrams -> crew diagrams. Very often it is two steps forward, one back until you reach the end, for example:

You might need to tweak the unit diagrams to rotate them for maintenance, which amends the timetable. Even if it is simply a unit swap at a location that alters the platform working it is still timetable data that needs changing.

In creating a traincrew diagram you might wish to amend which depot a train comes empty from, or goes back to. That affects all three functions. Keeping the crew with their train as much as possible, especially on short suburban workings, reduces the impact of disruption, so should be planned in at the start.

Traincrew are obviously an essential cost to any operator, so anything that minimises that cost is usually pursued to the end. Hence there are different new computer systems (as is being suggested for timetabling) that claim to assist with traincrew diagramming, but get the input wrong and the result can be disastrous.

 

[Bald Rick]

Well one can imagine, with the computing power at our disposal, we could build a model of the entire railway network and then automate timetable generation based on optimisation for various outputs that would be put into the system.
Would take quite a while to build the physical model and get it set up, but I imagine you could generate timetables in a matter of days, given a fast enough computer.

The network model is built in the Train Planning System and has been in use for some time.

A computer will only generate an answer according to the rules input to it. If it was that easy Network Rail’s Timetable Planning System would be more advanced than it is. It is not a quick win by any stretch and way beyond the tenure of any minister.

And a timetable is so much more than just a bunch of times. It is platform working, parallel moves over junctions, unit diagrams, rolling stock stock maintenance, traincrew diagrams, traincrew rosters, performance regimes, the list could go on.......and on........

Perfectly put.

 

[Ianno87]

That's precisely the kind of compartmentalised thinking that represents all that's wrong with the industry at the moment. The aim isn't "how can we run the best possible railway given £Xbn", it's "how do we make our life easier". Allegedly this new 'Fat Controller' set-up will avoid this, but until there are details it's difficult to be convinced.

I love references to the 'Fat Controller'; I suggest actually watching Thomas the Tank Engine to see how his style of perpetual fire-fighting micro-management short-termism is completely not what the railway needs.

I can see the benefits of it all being included but it will make the process a lot harder and longer to sort. You need a main timetable creating and then staff worked around that (with changes made if they have to be to accommodate staff and unit diagrams). Diagrams and the timetable have to be planned at around the same time but you do also need to keep some things separate to avoid NR doing all the work. TOCs can schedule things in a certain way to make things as efficient as possible. NR can and does plan some thing ineffectively and therefore costs more money in the end.
The more you give to one company, the more there is which could fail in one go.


The Thameslink stuff was due to a huge service change as well as issues with staff. Everyone has a part to play and NR can't just say 'we've done our bit, work with it', they do need to be able to adapt if train crew doesn't work but planning so much together means it can be less efficient and a railway costing more money isn't very good.

NR (as per the Network Code) do accept many, many 'spot bid' late changes from TOCs after D-26 to reflect optimisation for things like train crew.

 

[Bald Rick]

NR (as per the Network Code) do accept many, many 'spot bid' late changes from TOCs after D-26 to reflect optimisation for things like train crew.

Indeed, the entire timetable has been a giant spot bid* for three months!

*several hundred spot bids, in fact.

 

[The Planner]

Well you would pull out all the simulations that meet your decision criteria, and fit within the current requirements for rolling stock/staffing etc.
Chosing between fifty or a hundred cases of workable timetables is considerably less troublesome than generating one from scratch!

You could also apply a metric for "closeness" of train times to the previous one.

The decision criteria aren't defined in such a way that they can be put through an algorithm though. Have a look at Part D section 4.6 of the network code. Choosing between 50 or a 100 wouldn't be a short process either!

 

[HSTEd]

The decision criteria aren't defined in such a way that they can be put through an algorithm though. Have a look at Part D section 4.6 of the network code. Choosing between 50 or a 100 wouldn't be a short process either!

To mis-quote Napoleon "Je suis le déluge"
The network code might not survive this reorganisation in its current form.

Also - an easy way to select between them is to release all 50 options and see which one produces the fewest apopletic fits from stakeholder groups!

 

[route:oxford]

Well one can imagine, with the computing power at our disposal, we could build a model of the entire railway network and then automate timetable generation based on optimisation for various outputs that would be put into the system.
Would take quite a while to build the physical model and get it set up, but I imagine you could generate timetables in a matter of days, given a fast enough computer.

Lots of negative feedback to your comment, but it's absolutely accurate.

There is masses of computer power at our disposal. The average reader looking at this on their SmartDevice probably has more processing power under their domestic roof than every single machine that BR had put together when privatised.

The algorithms wouldn't start with nothing. They'd start with:-

The current working timetable.
Ticket sales and passenger volume data.
Historic track and unit performance data between every signal for as long a Network Rail have held it.

From then onwards, you start building rules and experiment with the machine learning.

Most importantly of all, it could be used for evidencing for investments in the Network and timetable refinements.

Let's just take two suggestions.

Divert the Highland Chieftain via Ladybank
Move the Stirling to London LNER service departure to 08:00 from Stirling or 07:30 from Perth

That would take weeks and weeks for Network Rail boffins to work out by hand, with their slide rules, on a signal by signal and junction by junction basis.

Whereas with machine learning, it'll instantly be able to supply details of potential benefits and conflicts.

(Has anyone told the First Minister and whoever is in charge of Wales that railways could return to centralised control in London?)

 

[Ianno87]

Lots of negative feedback to your comment, but it's absolutely accurate.

There is masses of computer power at our disposal. The average reader looking at this on their SmartDevice probably has more processing power under their domestic roof than every single machine that BR had put together when privatised.

The algorithms wouldn't start with nothing. They'd start with:-

The current working timetable.
Ticket sales and passenger volume data.
Historic track and unit performance data between every signal for as long a Network Rail have held it.

From then onwards, you start building rules and experiment with the machine learning.

There are thousands of rules and variables to consider. Many of which are subjective rather than absolute.

And the computer would need to be able ro identify when a conflict was unsolvable within acceptable parameters within these thousands of variables.

Most importantly of all, it could be used for evidencing for investments in the Network and timetable refinements.

Let's just take two suggestions.

Divert the Highland Chieftain via Ladybank
Move the Stirling to London LNER service departure to 08:00 from Stirling or 07:30 from Perth

That would take weeks and weeks for Network Rail boffins to work out by hand, with their slide rules, on a signal by signal and junction by junction basis.

For one train? No it wouldn't. 1 day of effort, tops. And nobody uses slide rules.

Whereas with machine learning, it'll instantly be able to supply details of potential benefits and conflicts.

If it was as easy as you suggest, it would've been invented by now.

 

[FQTV]

I’ve had quite a bit of behind-the-scenes exposure to how NATS (National Air Traffic Control Services) manages things in and over the U.K., but I’ve only experienced railway control at the passenger end.

NATS is quite mind-blowing in the way that it deals with global, local and human variables.

 

[route:oxford]

There are thousands of rules and variables to consider. Many of which are subjective rather than absolute.

That's fine, for the MRI data that I look at daily, there is always a subjective call. I can assure you there are millions of variables to consider when looking at a 4d flow of blood through major organs and the software is more than capable of handling it.

And the computer would need to be able ro identify when a conflict was unsolvable within acceptable parameters within these thousands of variables.

That's exactly what the humans do already. Don't the signallers manage to deliver too many trains into Canon Street every morning (might have the wrong station).

For one train? No it wouldn't. 1 day of effort, tops. And nobody uses slide rules.

So just to calculate the reflow of a single train on the Network would take a team a whole day. That's exactly why we need to develop this system.

If it was as easy as you suggest, it would've been invented by now.

Perhaps there are just too many people in the railway business who are saying "it can't be done". Probably the same one's who'll complain when we buy in a package in future from the another nation's operator that we should have done it here.

 

[Ianno87]

That's fine, for the MRI data that I look at daily, there is always a subjective call. I can assure you there are millions of variables to consider when looking at a 4d flow of blood through major organs and the software is more than capable of handling it.

That's one single subjective call.

A train timetable has thousands of subjective calls, the outcome of one directly affects a multitude of others. There is no single "correct" solution.

That's exactly what the humans do already. Don't the signallers manage to deliver too many trains into Canon Street every morning (might have the wrong station).

No, the signallers deliver a timetable plan that is (with a small handful of inevitable errors that do creep in) compliant and conflict free within the Timetable Planning Rules.

So just to calculate the reflow of a single train on the Network would take a team a whole day. That's exactly why we need to develop this system.

Who said anything about a whole team? It takes one, maybe two, skilled planners. Of which there are many.

Perhaps there are just too many people in the railway business who are saying "it can't be done". Probably the same one's who'll complain when we buy in a package in future from the another nation's operator that we should have done it here.

Maybe it can be done. But you need to actually understand the complexity of the problem, the number of variables and how they all interact and contradict each other first. First hand experience of actual train planning is invaluable in even starting to understand this.

 

[route:oxford]

A train timetable has thousands of subjective calls, the outcome of one directly affects a multitude of others. There is no single "correct" solution.

There are many, many subjective calls. Dying people are usually very complex cases with multi-morbidity. That's why we are data led. But with subjective consideration - Even when the data says "yes we can", some people just don't want any more procedures and to die at home.

But are you thinking about what you are typing? You are seeing it as an impossibility for IT and algorithms to cacluate what it takes two people an entire day to calculate and admit that what they come up with a solution that probably isn't "correct" anyway.

 

[liam456]

But are you thinking about what you are typing? You are seeing it as an impossibility for IT and algorithms to cacluate what it takes two people an entire day to calculate and admit that what they come up with a solution that probably isn't "correct" anyway.

Right! Perhaps there's a little chip on the shoulder of some who defend "the way it's always been done" but AI is capable of some very impressive stuff nowadays. Plus train planning (as complex and a dark art as it is) has more fixed outcomes and a more easily defined "win" than some of the application of neural networks/deep learning/AI in general.

EDIT: Perhaps someone from Milton Keynes could weigh in? Is such tech already in use and helping in train planning?

 

[Energy]

Right! Perhaps there's a little chip on the shoulder of some who defend "the way it's always been done" but AI is capable of some very impressive stuff nowadays. Plus train planning (as complex and a dark art as it is) has more fixed outcomes and a more easily defined "win" than some of the application of neural networks/deep learning/AI in general.

AI and machine learning could fairly easily do timetabling, it will still require instructions from human operators for how much time is required at stations etc but if it has the information it needs then it could easily do it.

 

[alistairlees]

There are many, many subjective calls. Dying people are usually very complex cases with multi-morbidity. That's why we are data led. But with subjective consideration - Even when the data says "yes we can", some people just don't want any more procedures and to die at home.

But are you thinking about what you are typing? You are seeing it as an impossibility for IT and algorithms to cacluate what it takes two people an entire day to calculate and admit that what they come up with a solution that probably isn't "correct" anyway.

I have been reading this thread and, in particular the part about timetable creation and modelling, with as much bemusement as you evidently have. A timetable has a very clearly defined set of inputs and rules. All very clearly lending themselves to the application of computing power to make the job both easier and for trade offs and options to be much better examined. And, in due course, for AI / machine learning to be applied. The challenge is to get the data structured and entered correctly in the first place; not for the computing power that’s required. On a super fast computer you could probably generate an entire GBTT in a few minutes, or hours at most.

It would be an obvious next step to feed this with actual train running performance, loading etc to input into modelling the next timetable (and to adjust any rules, or to identify any weaknesses, whether infrastructure, rolling stock or human, and to mitigate those).

I really hope this is the sort of thing that is actually being done.

Certainly the industry needs to find a way to get away from the vicious circle of:
- it takes years to do a timetable
- sometimes when a new timetable is implemented it doesn’t work, and passengers suffer (as does the revenue and reputation of the railway)
- but, as it takes years to do a new timetable, then passengers will have to suffer for years until we fix it

This completely lacks customer focus. Just because it’s the way it’s always been done definitely doesn’t mean it’s the right way.