Revenue management is important for carriers (e.g., airlines and railroads). In this talk we focus on cargo capacity management which has received far less attention in the literature than its passenger counterpart. More precisely, we focus on the problem of controlling booking accept/reject decisions: Given a limited capacity, accept a booking request or reject it to reserve capacity for potential future, higher revenue, bookings. We can formulate the problem as a finite-horizon stochastic dynamic program. The cost of fulfilling the accepted bookings, incurred at the end of the horizon, depends on the packing and routing of the cargo. This is a computationally challenging aspect as the latter are solutions to bin packing or vehicle routing problems. In this exploratory work, we propose to predict the solution costs to these discrete optimization problems using supervised learning. In turn, we use the predictions in an approximate dynamic programming algorithm to solve the booking control problem.
This talk is based on joint work with Justin Dumouchelle and Andrea Lodi (École Polytechnique Montréal).