A neural network approach to air cargo fleet assignment
This study explores the mathematical programming aspects of the air cargo fleet assignment problem for one international air cargo carrier - Korean Air - under given origin-destination (O-D) pairs, departure and arrival times, and frequencies. A pure cargo service is taken as the basis for this study, since such a service is not constrained by passenger route determinants and the schedule of a combination air carrier. The objectives of the study include: to identify the pure air cargo network representation of the combination air carriers; to develop and solve a conventional branch-and-bound mathematical programming model for optimising the assignment of aircraft to flight routes given a set of constraints, including aircraft fleet size, schedule balance, and `required through' constraints; to develop and solve the fleet assignment problem using a novel neural network optimisation modelling approach; to investigate methods of implementing the neural network model, and to analyse the performance of the model when compared with conventional solution methods; and finally to analyse the utility of the neural network model and identify how it may be used in the design and development of air cargo networks for combination air carriers like Korean Air. There are four main parts to the thesis: the first part outlines the schedule design process of an airline and some details of the fleet assignment problem are reviewed. The air cargo flight network is represented and the fleet assignment problem is formulated as a mixed integer programming problem of cost minimisation with various constraints. The complexity of the problem is discussed; the second part outlines the various techniques available to solve optimisation problems and neural network models are presented and discussed as a promising alternative solution method. Neural network applications in the transport field are reviewed and the neural network process for optimisation and for solving the general assignment problem are studied and presented; the third part incorporates the practical application of both the conventional fleet assignment problem solving method and the proposed neural network method to a combination airline's case - Korean Air. The detailed process of constructing a time line network and formulating a mathematical programming model are described and equivalent neural network models are formulated. The results from the two solution approaches are compared and evaluated; and the final part summarises the main findings, presents the significant conclusions, the contribution of the research is discussed and some recommendations for further research are presented. Overall, the conventional branch-and-bound optimisation model yielded plausible results which were demonstrably superior to those produced using the novel neural network optimisation models.