A combined forecasting and packing model to support air cargo revenue decisions



Air cargo revenue management departments are in charge of accepting or rejecting incoming booking requests. The overarching goal is to accept as many requests as possible, hence maximizing revenue, while limiting overbooking to ensure timely deliveries. In this paper, we present a novel combined forecasting and optimization model to assist the air cargo Revenue Management department of an airline. Using historical data from a partner airline, the forecasting block predicts the available cargo space in passenger aircraft (with a Long Short Term Memory network) and shipment dimensions (with a Multi-Layer Perceptron network), that are generally unknown at the time of the booking. On top of predicted values, probability distributions are also computed to be used as input for the optimization block. A Knapsack Problem is solved sequentially, using a heuristic based on the Extreme Points method, anytime a booking is received. A loading strategy of shipments, whose dimensions are computed using the aforementioned distributions and a user-defined confidence interval, into Unit Load Devices is determined, and the incoming booking is accepted if no other booking is offloaded in the process. The effectiveness, efficacy, and robustness of the model are tested on four case studies based on real booking data provided by partner airline.

Zoom: https://eur-nl.zoom.us/j/92878983187?pwd=NXp6aDFvWEhDNXJBeUtYUktTa2tnZz09
Meeting ID: 928 7898 3187
Passcode: 897793