Bus public transport algorithm

Question

I am working on an offline C# application that can find bus routes. I can extract the timetable/bus/route data. I am searching for the most simple solution that will work with

Answer 1

The problem you are working on is not a trivial task. So much so, that is has a name: the mixed integer nonlinear programming problem (MINLP). In the words of one author (Deb 1998):

"When formulated mathematically, the time scheduling problem becomes a mixed integer nonlinear programming problem (MINLP) having a large number of resource- and service-related constraints. Although attempts have been made in the past to find an optimal schedule of a simplified model using classical optimization techniques (Bookbinder & DCsilets, 1992; Kikuchi & Parameswaran, 1993), it is observed that this is an extremely difficult task even for a small transit network. The difficulty arises mainly because of the large number of variables and constraints, discrete nature of variables, and nonlinearities involved in the objective function and the constraints."

In Deb's paper he proposes a genetic algorithm.

Your other option would be to use simulation. Just to throw something out there you can try right away-- choose thousands of random routes that start at your origin, and fish out the ones that work reasonably well at getting to the destination.

Picture the algorithm like this: You are trying to find the quickest route from stop A to stop B, starting at a certain time. You hire 1,000 people and arm them with a quarter to flip. You tell them to flip the coin every time they have a chance to get on or off a bus. Heads, get off (or get on, if already off). Tails, stay on (or keep waiting, if off). They each have an index card to write down the choices they make as they go. You go to point B and wait for the first guy to show up and take his card.

Answer 2

Conceptually, you take the same basic algorithm for evaluating distance between A and B, but instead of distance, you should be evaluating time. Dijkstra can do both, if you give it the proper inputs.

You're used to seeing a map as a measure of distance. However, the same map can be a measure of time as well; all you need is to add data about average speed, and the time it takes to cover a particular distance of a particular road will shake itself out. You can even visualize the map in terms of time; routes that take longer will be longer. Dijkstra doesn't care which it's evaluating, really; it just cares about finding the continuous route with the lowest number, and whether that number represents length or time is immaterial.

To incorporate speed, naive algorithms simply use the daytime speed limit and assume you never have to stop while going from A to B; more advanced algorithms can incorporate information about time of day and traffic patterns (which will impact the average speed you travel on that road at that time), and whether a road is a freeway or surface street (and thus make educated guesses about time spent stopped at an intersection). What you use depends on what you have available, but a basic 4- or 5-layer time of day dimension should be adequate for all but the absolute most time-critical applications. For each direction of each road in your map, you need the average speed during morning rush, daytime, evening rush and night, possibly with lunchtime numbers as well. Once you have that, it's a relatively basic change to a Dijkstra algorithm to pass in a time of day and have it evaluate routes based on time.

Answer 3

read this:

Multi-Modal Route Planning. Master's thesis, Universität Karlsruhe (TH), Fakultät für Informatik, 2009. online available at http://i11www.ira.uka.de/extra/publications/p-mmrp-09.pdf

the section on railway routing also applies for bus routing.

The gist of it: the naive approach of expanding space and time into a single graph does not work for large networks. There are smarter solutions.

Answer 4

If you are interested in time information, then why not label the distance values on the graph edges using the time information instead of their physical distance from each other. That way you will search for the quickest route, instead of the shortest route. You can then use Dijkstra/A* in order to compute the your result.

I'm a little unclear on what you mean by time dependent. If you mean that that you need to answer queries of the form 'goes from x to y before 10am' then you can calculate what bus routes arrive before 10am, seems like a simple filter on the data. Then apply Dijkstra/A* to the data.

Answer 5

Try this as your data model.

Bus 1 goes to stops A, B and C. Bus 2 goes to stops B, D and E.

I would store a unique node based on both bus and stop, with distances to nodes being based on time. We would have node A1, B1, C1, B2, D2 and E2. In the special case of transfers apply the wait for the next bus as the distance between nodes. For example, if Bus 1 arrives at stop B 30 minutes before Bus 2, travel time from B1 to B2 is 30 minutes.

You should than be able to apply Dijkstra's Algorithm.

Answer 6

There is an extensive list of publications (30+) on public transport routing algorithms that has been compiled over time by contributors to the open-source (Java) OpenTripPlanner project here:

https://github.com/opentripplanner/OpenTripPlanner/wiki/RoutingBibliography

OpenTripPlanner is multi-modal routing engine that also includes bike and walk - from the above link:

This is a list of articles, dissertations, and books that have inspired and informed both the existing OTP routing engine and some ongoing experiments. Currently, OpenTripPlanner uses a single time-dependent (as opposed to time-expanded) graph that contains both street and transit networks. Walk-only and bicycle-only trips are generally planned using the A* algorithm with a Euclidean heuristic or contraction hierarchies. Walk+Transit or Bike+Transit trips are planned using a variant of the MOA* algorithm with epsilon-dominance for path pruning and the Tung-Chew heuristic (a graph providing a lower bound on aggregate weight) for queue ordering.

The routing bibliography above includes references for the following categories of algorithms and related work:

• Path Search Speedup Techniques
• Multi-objective Pareto Shortest Paths
• Resource-constrained Routing
• Contraction and Transfer Patterns
• Timetable-based routing
• ALT and Metric Embeddings
• Calibration and Implementation Details
• Post-Dijkstra Public Transit Routing

If you find something new that's not on the list, please feel free to add it to the wiki!

As far as other open-source public transport routing libraries - there is also the RRRR project by Bliksem Labs:

https://github.com/bliksemlabs/rrrr

From the above link:

RRRR (usually pronounced R4) is a C-language implementation of the RAPTOR public transit routing algorithm. It is the core routing component of the Bliksem journey planner and passenger information system. The goal of this project is to generate sets of Pareto-optimal itineraries over large geographic areas (e.g. BeNeLux or all of Europe), improving on the resource consumption and complexity of existing more flexible alternatives. The system should eventually support real-time vehicle/trip updates reflected in trip plans and be capable of running directly on a mobile device with no Internet connection.

Both OpenTripPlanner and RRRR use GTFS data.