The management of air cargo containers, or unit load devices (ULD), requires a large amount of manual labour. ULDs usually have to be broken down and the shipments therein have to be distributed among new ULDs bound to the shipments respective destinations. For build-up and break-down, a ULD has to be assigned a work station and a time window that is constrained by arrival and departure times and the available workforce. Mistakes in the planning step can severely disrupt the cargo flow since any wasted workforce capacity will miss later in the process. Planning is further complicated by usually incomplete information about measurements and weights of shipments and possible delays of incoming ULDs. The problem is currently solved with heuristics in industry applications. In this project, we will study which aspects of integer programming, robust optimization, stochastic methods and machine learning are best suited for the planning of air cargo ground handling. We suspect that the best performance will shift from machine learning/stochastical approaches to combinatorial optimization if we move from long-term to short-term planning.

Further, we will study in which way methods from both domains can be combined (e.g. prognosis, parameterization of the optimization model). For a realistic evaluation, we will also create a customizable simulation environment.