Enhancing efficiency in airside operations through prescriptive analytics

Abstract

Given the growth in air traffic demand and the highly competitive nature of the aviation industry, many airports face capacity challenges, leading to increasingly frequent and severe congestion. The aviation sector increasingly relies on predictive and optimisation techniques to fully utilise the massive data generated in daily operations. This thesis uses real-world data to provide informed decisions for airside operations by employing advanced prescriptive analytics methodologies, thereby improving efficiency and alleviating congestion. The first study introduces two prescriptive analytics approaches for the airport gate assignment problem, utilising historical data through machine learning (ML) techniques to enhance decision-making effectiveness and robustness. Supported by effective ML methods and scenario selection strategies, the estimate-then-optimise (ETO) approach delivers superior performance compared to other optimisation techniques. Additionally, we develop an efficient exact solution method, the Benders-based branch-and-cut (BBC) method, to effectively handle real-world scale test instances. This solution method demonstrates statistically significant improvements in computational performance over commercial solvers. The second study examines the aircraft sequencing and scheduling problem (ASSP) at a single-runway airport under uncertainty in aircraft arrival and departure times. We first introduce an ETO approach, utilising prediction results to drive the stochastic programming model for the ASSP. To address the suboptimal decision-making caused by prediction errors, we further propose an estimate-then-distributionally-robust-optimise (ETDRO) approach, which incorporates prediction results into a distributionally robust optimisation model for decision-making. Experimental results demonstrate that the ETDRO approach outperforms other optimisation techniques. Additionally, to effectively implement the ETDRO approach, we propose several exact and inexact decomposition methods. Extensive computational results show that our inexact decomposition method can provide optimal or near-optimal solutions for real-world scale test instances within a very short CPU time. In the third study, we focus on the prescriptive analytics of the multi-runway aircraft landing problem (MALP) with uncertain aircraft arrival times, aiming to design efficient and environmentally friendly landing operations. Following the ETO approach in prescriptive analytics, we employ ML techniques to estimate the distribution of uncertain arrival time based on historical data. An optimisation-enhanced learning-driven scenario generation (OLSG) method is used to generate scenarios that closely resemble actual scenarios based on the estimated distributions, thus preventing the subsequent optimisation from being affected by extreme scenarios and producing suboptimal decisions. Experimental results demonstrate the superior performance of the ETO approach supported by the OLSG method over other optimisation methods. Additionally, we propose a novel exact solution method called the stabilised branch-and-check (SBAC) method to solve the ETO approach for MALP efficiently. This method stabilises the master problem around a neighbourhood of a stable centre point, enabling the generation of strong Benders cuts. The results of computational experiments demonstrate that the proposed SBAC method achieves statistically significant improvements in CPU time compared to the benchmark methods. This thesis demonstrates the robustness and effectiveness of combining machine learning algorithms, optimisation techniques, uncertainty modelling, and advanced decomposition methods to address key operational challenges in airside operations. The results include significant improvements in operational efficiency, as well as economic and environmental benefits. In addition, the successful application of prescriptive analytics demonstrates the significant potential and advantages of data-driven decision-making in complex aviation operational environments. This suggests that these data-driven decision-making approaches can be introduced as innovative solutions to address various operational challenges within the aviation industry.