4 Presentations

• 
  10:30 AM - 10:55 AM

  Optimizing Empty ULD Repositioning Using Machine Learning and Optimization Techniques

  • Ali Barooni, presenter,
  • Frédéric Quesnel, GERAD & UQAM
  • François Soumis, GERAD & PolyMTL
  • Daniel Aloise, GERAD & PolyMTL

  Effective management of Empty Unit Load Devices (ULDs) repositioning is crucial for optimizing efficiency and reducing costs in aviation. Repositioning empty ULDs is expensive and constrained by limited flight capacity, while high demand variance makes distribution challenging. In this study, we tackle imbalanced ULD distribution across airports by developing a data-driven approach to optimize their movement.

  Our approach has two stages. First, we forecast ULD demand per flight using features such as aircraft type and route characteristics. We initially applied regression-based models—including XGBoost and a Zero-Inflated Poisson (ZIP) model—to predict ULD requirements. To boost accuracy, we then incorporated time series techniques using Prophet and other methods to capture temporal dependencies in each airport. Second, we built an optimization model to determine the optimal movement of empty ULDs across the network, ensuring cost-effective repositioning while considering operational constraints.

  A key contribution of our work is leveraging real-world airline data to enhance the reliability and relevance of our findings. Although similar optimization methods have been used in container shipping, their application in aviation for ULD management is emerging. Our results offer valuable insights for airlines aiming to reduce empty repositioning costs, improve fleet utilization, and minimize unnecessary transfers through data-driven decision-making.

• 
  10:55 AM - 11:20 AM

  Learning-to-preprocess for the crew rostering problem

  • Philippe Racette, presenter, Polytechnique Montréal
  • Frédéric Quesnel, Université du Québec à Montréal
  • Andrea Lodi, CERC, Polytechnique Montréal, Montréal, Canada and Jacobs Technion-Cornell Institute, Cornell Tech and Technion - IIT, New York, USA
  • Francois Soumis, GERAD et Polytechnique

  Machine learning (ML) has become increasingly popular to improve the resolution of combinatorial
  optimization problems, although the factors involved in the successful interaction between the two are
  not always clear. In this talk, we present the case study of the crew rostering problem (CRP), a complex
  task aiming at creating monthly schedules for pilots while accounting for their preferences. We
  consider different forms of learning-to-preprocess, an ML family of methods whose goal is to better
  configure a traditional optimization solver before resolution starts. We find applications for the CRP
  and its accelerated resolution through windowing with ML-generated initial solutions. However, the
  initial solutions are strongly incompatible with optimal rosters from the point of view of dynamic
  constraint aggregation despite the implementation of several methods to enhance their quality. Based
  on these results and a larger discussion of the literature, we conclude that unless a special structure can
  be exploited for a given problem, learning-to-preprocess can improve upon the state of the art under the
  condition that the information provided as input to the solver is not subject to stringent requirements,
  such as the imitation of an optimal solution. Future research directions are then discussed.

• 
  11:20 AM - 11:45 AM

  Short-haul flight bans from the network perspective: a benevolent policy or an ineffectual mistake?

  • Ye Xinjun, presenter, HEC Montreal

  Amid the ongoing or impending bans on short-haul flights in several localities where cleaner transportation modes such as railways are available, this paper explores the change in air and railway ridership in response to such bans. We investigate the behavioral, monetary and environmental ramifications of the ban in a hub-centered transit network, starting from the case where one long-haul international and one short-haul domestic leg could connect at the hub. The Bertrand model of price competition is proposed, where each operator maximizes their respective overall revenue, with considerations for substitution effects. It is found that the short-haul flight bans would sometimes increase the emissions of the long-haul trip, due to connecting effects. The criteria for decreased total emissions are analyzed. We provide numerical insights using data based on the Montréal-Paris-Lyon corridor. Eventually, meticulous planning and cooperation of stakeholders is advised to ensure balanced and salutary effects on society.

• 
  11:45 AM - 12:10 PM

  Improving Regularity in the Crew Pairing Problem: A Hybrid Bonus-Based and MIP Approach

  • Mohamed Faouzi Benammour, presenter, Gerad
  • Frédéric Quesnel, Polytechnique Montréal
  • Francois Soumis, GERAD et Polytechnique
  • Benoit Rochefort, GERAD

  The Crew Pairing Problem (CPP) involves constructing feasible pairings (sequences of flights,
  connections, and rest periods) for airline crew while minimizing operational costs and
  ensuring full flight coverage. However, beyond cost minimization, regularity has become
  a crucial objective for many airlines, as it enhances operational stability and reduces
  indirect costs associated with hotel accommodations, transportation logistics, and crew
  training. Traditional CPP optimization methods primarily focus on cost reduction but
  often fail to effectively incorporate regularity, as they lack a structured mechanism to
  balance both objectives.
  This study proposes several approaches to enhance regularity in the CPP. The first is a
  bonus-based approach that identifies pairings with high repetition potential in the initial
  solution. The CPP is then re-solved using a branch-and-price algorithm to encourage the
  selection of these repeatable pairings. The second approach reoptimizes the initial
  solution by solving a Mixed-Integer Programming (MIP) model that explicitly maximizes
  regularity. Finally, both approaches are applied sequentially to further enhance
  regularity while maintaining cost efficiency, forming a hybrid method.
  Computational experiments conducted on datasets from a major Asian airline
  demonstrate the effectiveness of the proposed methods. The results show that they
  generate highly regular solutions with minimal cost deviations, ensuring practical
  applicability in airline crew pairing optimization.