4 présentations

• 
  15h30 - 15h55

  The vehicle routing problem with driver scheduling

  • Razieh Mousavi, prés., Université Laval
  • Jean-François Côté, Université Laval
  • Maryam Darvish, Université Laval

  This talk addresses the Vehicle Routing Problem with Driver Scheduling (VRPDS), an integration of the Vehicle Routing Problem with Time Windows (VRPTW) and personal scheduling problem. In the VRPDS, a company has to fulfill a set of known requests using a team of drivers over a one-day planning horizon. In this problem, we relax the commonly used assumption of fixed driver availability where drivers have their shifts.

  Each driver can have a preferred availability period, and they are given a schedule with a start and end time that respects their availability. It is assumed that drivers operate identical vehicles and must be allocated to the routes to do the deliveries. The requests not served by the fleet are given to a third-party logistics company.

  The problem is formulated as a deterministic integrated VRPTW and shift scheduling problem. Instances are generated randomly based on the literature, and a heuristic algorithm is developed to solve the problem.

• 
  15h55 - 16h20

  Reinforcement learning and column generation for optimizing log truck scheduling: An integrated approach

  • Abdelhakim Abdellaoui, prés., Polytechnique Montréal
  • Issmail El Hallaoui, GERAD, Polytechnique Montréal
  • Loubna Benabbou, UQAR
  • François Aubé, CanmetÉnergie

  This work introduces an enhanced mathematical programming model utilizing a time-space network representation, tailored to address real-time transportation challenges in the forestry sector. Our study aims to simplify transportation modeling in forestry, including unforeseen events, through a novel approach. Initially, we focus on the mathematical modeling of the problem. We then concentrate on learning key parameters, such as the time discretization interval, to facilitate solving large-scale instances efficiently. Our approach integrates column generation techniques with reinforcement learning and local search techniques to develop a robust solver for multi-attribute, real-world data. The empirical evaluation of our model was conducted using data from a Canadian forestry company. The results demonstrate the approach's effectiveness in simplifying the problem formulation, achieving optimal solutions within reasonable computational times for real-world instances. These findings underscore the model's potential in enhancing transportation operations, highlighting its practical benefits in forestry logistics.

• 
  16h20 - 16h45

  A Column-Generation Algorithm for the Intercity Electric Coach Scheduling Problem

  • Malte Billen, prés., The University of Edinburgh
  • Jörg Kalcsics, The University of Edinburgh
  • Sergio Garcia, The University of Edinburgh
  • Goncalo Dos Reis, The University of Edinburgh
  • Adrian Grütter, Ember Core Ltd

  Governments globally are incentivising zero-emission vehicle sales and charging infrastructure development to cut transportation emissions. These incentives, along with higher market availability, lower fuel cost and extended battery range, make electric coaches viable for intercity public transportation. In collaboration with Ember Core Ltd., the first intercity operator in the United Kingdom to use an all-electric fleet, we have conducted research on maximising the utility of their electric coaches and charging infrastructure while ensuring that schedules fulfil the operational restrictions of intercity electric transportation. As a result, we introduce the Intercity Electric Coach Scheduling Problem along with a column generation algorithm that uses a variable fixing heuristic to find integer solutions. The algorithm is embedded in an iterative framework to generate schedules for a long planning horizon. It provides quality schedules that successfully meet Ember’s real-world requirements. In the forthcoming presentation, we will discuss the challenges of sparse charging networks, regular maintenance scheduling and overnight timetables that make intercity electric coach scheduling a difficult problem to solve. However, we also explain how we can exploit these characteristics and provide a comprehensive analysis of the algorithm’s solution quality and performance.

• 
  16h45 - 17h10

  The Electric Bus Rostering and Charging Scheduling Problem with Uncertain Energy Consumption: a Two-Stage Stochastic Programming Approach

  • Kyra D'Ignazio, prés., HEC Montreal
  • Jorge Mendoza, HEC Montréal

  Electric vehicles are one of the promising technologies that are being used in the transportation sector to help reduce the negative impacts associated with traditional technologies. In order to integrate electric vehicles into public transportation, this study focuses on the examination of how energy uncertainty in electric battery consumption impacts operating costs associated with charging decisions. This research formulates the energy uncertainty challenge as a two-stage stochastic programming model by employing mixed-integer linear programming. The model optimizes bus route assignment and charging decisions such as when to charge, while accounting for the potential deviations in energy requirements that may require a recourse action in certain scenarios. The results establish a correlation between the level of uncertainty in consumption and the difficulty to solve the problem (i.e., CPU time). This research contributes to the realm of electric vehicle fleet management by providing a comprehensive methodology to optimize bus route assignments and charging protocols considering the real-world variability in energy consumption.