Help: Setting multiple SoftUpperBound for different vehicles

[jayadevc]

I have modeled a VRP with capacity and time window constraints which are working perfectly.
Now I need to add cost of vehicles depending on the total distance travelled by the vehicle.
The cost of a vehicle consists of different slabs depending on the distance traveled as shown below.

-- Vehicle_1 --
Slab 1. Fixed Cost: 10000
Slab 2. 0 to 100 km / Cost: 5000
Slab 3. 101 to 250 km / Cost: 7000
Slab 4. >250 km / Cost: 100/km

Suppose the distance traveled by vehicle_1 is 270 km, cost will be 24000 (10000 + 5000 + 7000 + 20*100)
For each vehicle cost slab will be different.

I have achieved first and forth slabs using following method -
distanceDimension.SetFixedCostOfVehicle(10000, vehicle);
distanceDimension.SetEndCumulVarSoftUpperBound(vehicle, 250, 100);

For achieving 2nd and 3rd slab I tried using SetEndCumulVarSoftUpperBound multiple times with different upper bound and coefficient, but only the last SetEndCumulVarSoftUpperBound
is working.

Can someone suggest a way to achieve this?

or-tools version: 6.5
lang: C#

[GuyBenhaim]

You could implement 3 separate dimensions, force their accumulated values to be offset properly, and apply an upper bound on each of these. ?

[jayadevc]

Good suggestion. Also when I tried to implement SetCumulVarSoftUpperBound directly into my code, the solver uses upper bound as a hard constraint rather than finding a solution with minimized cost . The code I used is shown below.

for (int vehicle = 0; vehicle < number_of_vehicles; ++vehicle)
{
IntExpr expr = solver.MakeSemiContinuousExpr(solver.MakeSum(distDim.CumulVar(model.End(vehicle)), -upper_bound), fixed_cost, per_km_cost); 
IntVar costVar = expr.Var();
model.AddVariableMinimizedByFinalizer(costVar);
}

Did I go wrong somewhere?

[xomaczar]

@jayadevc I've got a similar problem, have you been able to make any progress on yours?

[lperron]

There are piecewise linear cost, but only in C++.

[bmenendez]

Hi! I would also be interested in using piecewise linear functions in Python for a routing problem. Does the last messages of @Mizux about modifying the milestones mean that it will be available through 9.1 version?

Thank you very much for your efforts on doing or-tools an excelent way of solving optimization problems.

[jayadevc]

@Mizux Will piecewise linear cost function available in 9.1 for C# ?

Its much awaited feature for which we have been waiting for past 2+ years.

Thanks for all your efforts...