Trip Generation and Distribution

[The Archcontroller]

Wooo! First Post!

Just discovered Transport Empire, and I think it's great! From what I've seen, there's a lot of enthusiasm here, and I think a new transport game is in order.

I don't know if this has been suggested before, or even implemented in another game (Traffic Giant, I'm looking at you!), although I know it's not a part of the TTD economics engine: Realistic trip distribution modeling (weeelllll, maybe not THAT realistic). Basically, instead of treating passengers as 'goods' which well up from a city and can be transported to any onther city, the game should assign 'destinations'. In other words, each map automatically generates a matrix which tabulates the maximum number of passengers in a given time period (ie, a month) travelling from each city to each other city (the maximum potential number of trips).

This way, passengers will only board vehicles which travel to their desired destination, thus allowing a more realistic approach to route-designing (ie, like train lines instead of point-to-point).

I know that it's possible to generate such a matrix using what is known as the Gravity Model of Trip Distribution, but my question is: is this something we want in the game? (I say 'Yes!') Let me know what you developers and fans think!

[Steve]

I have always imagined a system where passengers arrive at a station with a destination and as you say, only board transport that will take them to or closer to their target. We would also consider route lengths, number of switches and such, when evaluating if passengers would take that route.
This should allow airports to be placed way outside of towns, just like in real life. The passengers are then more than happy to take the bus/train/tram to the airport, which will get them across the large distances and to their destination. Passenger destinations would need to be populated fairly randomly, with some preference on the size of the city they are heading or if we got really serious, each city would have some sort of popularity rating, dependent on size, scenery (beach etc), transport links.
This would allow nice sharing of competitors infrastructure too, you could provide the train link to your competitors airport, with each leg of the journey earning the appropiate player money.

[The Archcontroller]

I kinda see what you're saying with respect to route lenghts and etc: you're basically referring to the 'mode choice' part of the trip distribution. I was more focussing on the maximum number of passengers per month that travel from one city to the other, which would involve an impedance function based on the distance btwn the cities. My approach to implementing the actual mode choice (ie, which transport company or transport service passengers use) was to set up 'routes' or 'lines' which trains are assigned to, which then have a time-based utility rating. Obviously, though, we cannot have passengers only choose the fastest route; actual conditions should be spread based on the economy of the city. For example, a 'rich' city has lots of business travellers which can afford the fastest trains, but a less-prosperous one would prefer a cheaper one. Alternatively (or, in addition to), access from the city could (should?) be a facotr in the choice: a well-connected service (one with feeder lines) would do better than one without.

[prissi]

I think many of these aspects could be done be consideration two goods "first class passengers" and "normal passengers".

However, from my simutrans experience, a relatively simple passenger model (with fixed destinations per passenger) is good enough for nearly all gaming purposes. Building a network is hard enough. And when it is built, then keeping it running wiht the increasing amount of passengers (due to connection in the network) is challenge for a very long time.

In simutrans the destination selection is based upon city size and level of a building. In my OpenTTD-patch I just used the number of passengers started a station last time, which of course is similar. The newest patch has a slightly different system, which is more complex (imho too complex to handle millions); but in the end it boils down to the same: Even a total random selection would prefer larger towns, since more stops are intown (assuming complete coverage of a town.)

To check this I just suggest to try simutrans or the passenger destination patch.

(And there is already a second network additional to normal passengers: mail, which is different, like less goes to tourist attractions but much more returns and so on.)

[The Archcontroller]

Hmmm.....I'm not familliar with the patch you described, nor was I aware that Simutrans had this sort of capability (I tried to get into Simutrans, but never liked the track-building methods in it...). As to the method you've described to me that it uses, it looks like some kind of cross between a disaggregate and an aggregate model of distribution, which is simillar to what I'm proposing anyways.

In any case, since we have the opportunity to start from scratch, the model I'm suggesting is based on an aggregate model used in real transportation planning, giving it a bit more versimilitude. Basically, the model assigns a number of productions and a number of attractions to each municipal body, which is the sum of the individual production and attraction values which would be assigned to each building. A matrix is then generated for the entire map which contains the potential number of passengers which want/may travel from each city to each other city.

The only disadvantage of this model is that it 'only' works for interurban travel - as it applies to the whole city, not just what is in the catchment area of the station. In other words, in order for it to work, stations would service the entire municipality. My question to the community-at-large is: what do you think about stations servicing cities and not just what is next to them?.

Obviously, this wouldn't be the whole story - what about really large cities, or competing companies, etc, etc. The model I'm proposing here is simply a distribution model - the actual number of passengers waiting at a station would be dependant on a number of other factors, which I submit is a different discussion entirely.

[ohlidalp]

what do you think about stations servicing cities...?.

Though I'm an *TTD player, I think it's a good idea. Catchment-defined stations are good for cargo-like passengers, but for 'thinking' passengers, it would be irrealistic - people won't ignore a station in their town just because it's a few blocks away (usual in TTD ). This would especially apply to airports, where in reality nobody expects them to be in the neigbourhood (in fact, they expect them to be far enough because of noise - that could be worked-into TE too ).

...airports to be placed way outside of towns, just like in real life. The passengers are then more than happy to take the bus/train/tram to the airport...

... it 'only' works for interurban travel...

I like Steve's plan, but does the trip distribution model solve this? I guess so, but I'm not familiar with it at all, so I'm asking.

[Steve]

As is pointed out above, I'm not sure having an entire city as a single entity will work for productions. For attractions it will be fine, real life passengers would be attracted to a city if it can offer A LOT to them.
But similar to the airport model, lets say you live on the outside of a city, with the intercity train station in the middle. It's unrealistic to warp the passenger from their house to the train station, we'd prefer them to catch one of our nice buses.

But then this gets very tricky. How do we model productions?
Around a station: possible, then we'd need to add it into the matrix for finding routes between parts of the map. Two stations in the same area, are passengers split? Do they appear in both stations? Will an area of heavy population of stations become hell due to varying catchment areas, passenger splits and route finding?
By city: we discuss above.

By district: Lets say the city was automatically segregated into several areas. Those familiar to London could relate it to postal area codes for instance. It could then be implied that travel within a district is instant and we pool production from a district. This production would then be on offer to any station in the district, split depending on quality of service, much like in Transport Tycoon. Players would then aim to link up these districts to provide passenger travel.
These districts wouldn't be huge, but would be significant and of course, borders would need to change over time as the city did. They would probably relate somewhat to the style of buildings in each area. Assuming our city growth model knows the concept of residential, commercial and industrial (as in Sim City), and would create little areas of each, these would form ideal districts.

[The Archcontroller]

First of all, thank you for your comments - I guess I didn't explain it well at first, and there have been a few things which I didn’t consider. Also, I haven't defined some of the terms I am using.

Trip Distribution: The demand for travel from one municipality to another
Trip Generation: 'Where' the passengers come from, how they get to the station, etc.
Mode Choice: How passengers decide between airports vs. trains, or Player 1 vs. Player 2.

These steps require, to some degree, seperate modeling. At first, I was describing the first two steps, but I've realized that I can only confidently describe the first step right now.

TRIP DISTRIBUTION MODEL
You're confusing the trip distribution with the mode choice. The gravity model I've been talking about is visible to players, to some degree, which they can then use to design their routes. It is in existence before the route would be created. Productions and attractions are characteristics of a population, and would correlate with but not dictate exactly how many passengers travel on your routes.

Let’s take an example:

A player wants to create a passenger network in a region. He opens us a ‘network map’ (it doesn’t have to be a map) which shows arrows from each city to each other city (maybe laid over the landscape directly as a graphical effect). The thickness of the arrows represents the demand of travel between the two cities it connects (and it would also be labelled with the ‘true’ value). He creates his network based on this information.

The rest of it - the actual trip generation, and mode choice - I don't yet have in a polished form. I have an idea of how to do it, based on a 'uniform' distribution of the trips over the city and relative probabilities, but please let me write it up before commenting.

[The Archcontroller]

Done! Download it, check it out.

Let me know what you think.

[Steve]

Very good.

Some points:

• Routes are good, I agree with the concept. It makes the whole idea of routefinding easier, junction handling, etc. It also allows you to timetable trains in some fashion, to ensure they do not get bunched together. The game can assess the length of the route and get trains to wait until leaving the station.
• I think design mode has been accepted for quite a while. The only wonders I have are with the time, if time flows freely, what happens when a town starts to build on your ghost track? The subsidies are certainly an interesting idea that I hadn't considered. But remember, there's no need to build an entire line at one, in a single build mode instance.
• You say potential passengers can't be seen by the player.. why not? That would surely help the player find routes to build on, rather than randomly building and hoping. Unless you want them to pay for market research.
• Catchment is interesting, it was always a pain in TTD and Loco, where it was hard to build inside cities and see what's happening. But if it was possible in Tempire to do it properly, theres no reason we should cheat it. Modelling Taxis, cars etc down there would also be an interesting challenge.
• I do very much like the idea of using a real model.. even if I don't fully understand it

Thanks

[The Archcontroller]

Very good.

Thanks!

To answer your concerns:

what happens when a town starts to build on your ghost track?

Well, the beauty of design mode is that you're free to make neccessary changes. But, I guess if you're building a station, it'd be difficult to redesign the entrance. My only solution to this is maybe players can 'submit' designs to towns prior to construction approval? Like, a 'saved' design which becomes visible to other players and prevents them and towns from building on it? I say, it depends on the coder, and whether he/she is willing to put the time into it.

But remember, there's no need to build an entire line at one, in a single build mode instance.

Exactly.

You say potential passengers can't be seen by the player.. why not?

They can...and they can't. As I stated earlier, the maximum potential passengers which travel from each town-pair can be seen by players, as it represents general knowledge of how many people travel from place to place. However, the specific potentials used in the Trip Generation stage represents individual passengers who may or may not use the system. Like in real life, you can only see those who choose to use your service (ie, the actual passengers).

Modelling Taxis, cars etc down there would also be an interesting challenge

Yes. Carreers have been built on modelling automobile transport patterns. Call me back in three years, and I can charge you $60,000 (at least!) to do it .