"In your current solution, the vehicle has to fulfill one paired orders (a trip) before starting the next one. Therefore, the transit time for the paired orders is the shortest time between the first and second order in the pair. You can use the current transit time as the MaxTransitTime, which essentially is like a fixed transit time in this case."
Because I simplify the public transport network it's not only the travel between the order pairs. I removed all intermediate stops so the travel time between the order pairs is much shorter than the real travel time of the trip. So I use the restricted time windows to model the real travel time. The picture shows the route in reality with a lot of deviations, so when I model only the first and the last stop I will create a shorter route on the direct connection.
A possibility would be to model all intermediate stops but this would take a lot of effort, and I had to put a lot more data in the model.
[ATTACH=CONFIG]22149[/ATTACH]
One result of this simplified bus routing and the hard restrictions is that the busses spend a lot on waiting time:
[ATTACH=CONFIG]22152[/ATTACH]
"In addition, please make sure the MaxViolationTime1 and MaxViolationTime2 are set to null for all orders so all time windows are soft restrictions. In the layer advanced settings, please choose the time window violations importance to low so there will be less penalty for time window violations. If you are free to set any time windows for those trips, I think you can remove all time windows to solve it. Solver will return you a lower cost solution with potentially less vehicles."
I think I have to use hard restrictions because it's mainly school transport I model. We have some regulations that the whole travelling time (walking time to stop, time in the bus, waiting time at school before the class begins) is limited e.g. to 80 minutes and the start of school is fixed (is this scenario).
I think I do something with the VRP for which the tool wasn't made for, and I want to do a little more.
Update:
Probably a solution would be to add the waiting time which results from the strict time windows as a service time in the first or last stop of the order pair. Afterwards I can open the time window to model some scenarios.