This is a challenge from my Bachelors of International Aviation Management from 2017. The goal was to develop a flight plan for DHL out of their (at the time) new hub Leipzig. This comprised planning for the transport of packages on 173x172 OnDs via train and Road Feeder Services while adhering to time constraints for pickup, delivery, slots @LEJ, etc. By that time, we completed the project without numerical optimization methods (as expected by the university).
However, I quickly realized that this is nothing but an optimization problem with many constraints, minimizing cost. Therefore it must be able to solve it systematically. This is the long term aim of this project. In this context, I wrote a small program in 2018 using a Google API to measure the street distances for all ca. 30000 OnDs.
I have not worked on this problem for a longer time, and it is also not high priority as this is more of an operations research problem. However, I like modelling it and I plan on coming back to it once I have a better foundation in Data Science & AI methods.
Overall, we're minimizing cost. Basically, for facilitation of the problem, I suggest a four step approach:
Trucking is nearly always cheaper than sending a plane. We truck directly whatever the time constrains allow to truck. The network is not optimized in this step yet.
2. Aggregate remaining freight volumes to 'Business Centers' (Airports with greatest time constraints) wherever possible
Also multileg flights are an option
Look, which flying OnDs still have enough capacity to take further freight and make a direct road feeder unnecessary.
For the flight cost calculation, we need to rely on an approximation as we used a separate program for the calculation of the cost. We have the exact cost for all flights to LEJ, but sometimes multileg flights make sense. In this case we have to rely on cost estimates for the second leg. I am currently solving this by a linear regression per A/C-type with distance as the dependent variable. The same applies to flight times.
The first point is solved, we have the distances and we truck whatever is possible. The next step would now be to check whether 2. and 3. can be combined and be formulated as a minimization problem.
I have not worked on this problem for a longer time, and it is also not high priority as this is more of an operations research problem. However I plan on coming back to it once I have a better foundation in Data Science & AI methods.