feat(drt): check for maxWaitTime being lower than maxTravelTimeBeta

[marecabo]

I would suggest to add a constraint "DrtOptimizationConstraintsSetImpl.maxTravelTimeBeta > DrtOptimizationConstraintsSet.maxWaitTime" in DrtOptimizationConstraintsSetImpl.

• DrtOptimizationConstraintsSetImpl defines an upper bound for the travel time of a DRT trip with its maxTravelTimeAlpha and maxTravelTimeBeta values.
  

  matsim-libs/contribs/drt/src/main/java/org/matsim/contrib/drt/optimizer/constraints/DrtOptimizationConstraintsSetImpl.java

  Lines 10 to 22 in f466c47

  	@Parameter
  	@Comment("Defines the slope of the maxTravelTime estimation function (optimisation constraint), i.e. "
  	+ "min(unsharedRideTime + maxAbsoluteDetour, maxTravelTimeAlpha * unsharedRideTime + maxTravelTimeBeta). "
  	+ "Alpha should not be smaller than 1.")
  	@DecimalMin("1.0")
  	private double maxTravelTimeAlpha = Double.NaN;// [-]
  	@Parameter
  	@Comment("Defines the shift of the maxTravelTime estimation function (optimisation constraint), i.e. "
  	+ "min(unsharedRideTime + maxAbsoluteDetour, maxTravelTimeAlpha * unsharedRideTime + maxTravelTimeBeta). "
  	+ "Beta should not be smaller than 0.")
  	@PositiveOrZero
  	private double maxTravelTimeBeta = Double.NaN;// [s]

• These are used to compute a max travel time:
  

  matsim-libs/contribs/drt/src/main/java/org/matsim/contrib/drt/routing/DefaultDrtRouteConstraintsCalculator.java

  Line 45 in f466c47

  double maxTravelTime = defaultSet.getMaxTravelTimeAlpha() * unsharedRideTime + defaultSet.getMaxTravelTimeBeta();

• The resulting DrtRouteConstraint.maxTravelDuration is then used in the DrtRequest as follows:
  

  matsim-libs/contribs/drt/src/main/java/org/matsim/contrib/drt/passenger/DrtRequest.java

  Lines 60 to 61 in f466c47

  	latestStartTime = earliestStartTime + builder.constraints.maxWaitDuration();
  	latestArrivalTime = earliestStartTime + builder.constraints.maxTravelDuration();

  
  Obviously, it must hold that latestStartTime < latestArrivalTime.

• Tracking this back, I would guess, that maxTravelTimeBeta must be always greater or equal to maxWaitTime.

Do you have any thoughts on this? Maybe I am missing something?

Edit: The first commit is the actual change, all subsequent commits are the necessary adaption of tests.

[steffenaxer]

LGTM

[nkuehnel]

I'm not sure I agree. There can be scenarios where one would have max wait > beta travel time. In these cases, the max wait time may only come into effect for long distances, compare the following:

In the right graph max wait is higher than beta, however, it can only come into effect for long trips (when alpha >1). In this case, a trip can be picked up late, but must be quite direct.

As such, latest start and latest arrival can be defined independently, only insertions that comply with both are valid.

[nkuehnel]

Sorry the above graph assumed a minimum allowed detour, without that there is no shift to the right, but the argument is the same:

[nkuehnel]

I would make it a warning that waiting time may not always be fully effective only.

[nkuehnel]

Simple example for the argument above:

given max wait of 10 mins and beta = 5min and alpha = 1.3 (30% detour)

Given a trip with a direct ride time of 30 min.
Then max travel time is 5 + 1.3 * 30 = 44 min.

This means latest arrival at departure time = 0:00 is 0:44.

This means, a direct ride without further detours lasting 30 mins does still have room for 14 mins of waiting time, which would then be capped by 10 mins max waiting time.