Solving Aircraft Routing Optimisation Model with Operational Research and Deep Learning Methods

Abstract

This seminar addresses the aircraft routing (AR) problem by framing it as a set covering problem. It utilises a hybrid method that integrates graph neural networks (GNNs) with mathematical programming. A deep learning model based on GNNs produces high-quality initial solutions for the AR optimisation task, minimising computational complexity through an attention mechanism, residual connections, and distributed caching. Simulation results reveal the algorithm’s effectiveness, achieving an initial solution with a minimum gap rate of 6.73%. The solution time for the AR issue is cut by up to 50% compared to a standard Gurobi solver. This proposed model is particularly beneficial for large-scale optimisation challenges, quickly delivering initial feasible solutions and enhancing convergence speed. Additional model features, such as multithreading and caching mechanisms, further boost computing performance for datasets containing up to 7,000 variables. This seminar underscores the scalability and efficiency of deep learning models in addressing complex, large-scale optimisation problems.

Speaker

Prof. Cheng-Lung (Richard) Wu, an Associate Professor at the University of New South Wales, specialises in airline operations management, airport terminal planning, passenger behaviour analysis, and big data analytics. His recent projects include optimising fuel policies and maintenance scheduling for Qantas, developing AI models for Virgin Australia’s frequent flyer program, and examining passengers’ retail spending behaviours at the airport. His work has been extensively published in leading journals such as Transportation Science and Tourism Management. Major organisations, including International Air Transport Association(IATA), airlines, and airports, have embraced his research findings. His recent projects focus on deep learning models in optimisation, reinforcement learning models in scheduling, and autonomous drone mission control.