On Tue, Jun 3, 2008 at 6:21 PM, Stefan de Konink <[EMAIL PROTECTED]> wrote: > Basically for every way/node a cost function based on historical data. > If you route over the node you use Dijkstra/A* and let the cost or > heuristic aspect to take in account the time you will arrive on that > place. (With a nice error bound)
This does not explain the following observation : > camaj schreef: > a two hour journey in rush hour can often be a 10 minute one at 4am Before we can think of algorithms, we must have a mathematical model. Clearly travel time / average speed does not have a normal (Gaussian) distribution. Traffic jams may also be explained by bending the laws of fluid dynamics : If one lane of a three lane highway is blocked somewhere, we can use Bernoulli's law to predict what the cars must do to prevent a traffic jam from forming, namely they must travel faster or closer together at that point. (Yet curiosity cause some idoits to slow down) Journey time prediction can also involve risk management : If you have a list of appointments then the planner may need to tell you that you need to skip the first one because chances of making is too small. I did a little bit of mathematical modeling for the local railway company and it became clear that the smallest failure can cause massive losses for them and / or their clients : A simple IP packet being dropped and timed out can cause a complete train to be delayed. Therefore the software must keep on retrying in an endless loop. Similarly it's important for the governments to do small things well, like clear up accidents quickly, so that people can arrive on time and be productive. But can we really predict when a longer route will be faster ? And do it better than experienced driver ? _______________________________________________ Routing mailing list [email protected] http://lists.openstreetmap.org/cgi-bin/mailman/listinfo/routing
