TA13 - Concours d'affiches / Poster Competition

This study develops a predict-then-optimize framework for the Green Vehicle Routing Problem with Time Windows under time-varying traffic. Gradient boosting models trained on Caltrans PeMS data forecast hourly speeds, embedded within a multi-stage optimization model that selects routes and departure times to reduce CO2 emissions relative to static routing approaches.

A surgical episode traverses a subset of perioperative care stages before completing the care pathway, imposing distinct clinical requirements and generating uncertain demand for resources across these units. In this paper, we model and optimize surgical care pathways for same-say patients within a strategic network of collaborating hospitals in the University Health Network (UHN) in the City of Toronto, to minimize the risk-adjusted network-wide maximal completion time. We capture the inherent uncertainty in surgical pathways using the coherent risk measure Conditional Value-at-Risk (CVaR) and, more broadly, other coherent risk measures. We novelly incorporate these uncertainties into a mixed-integer programming (MIP) and a constraint programming (CP) model. Because these formulations integrate numerous assignment, sequencing, and scheduling decisions, they become computationally intractable at realistic scales, albeit with different computational behaviors. To address this challenge, we exploit the decomposable structure of the problem to develop two-level and multi-level classical and logic-based Benders decomposition algorithms that leverage the complementary strengths of CP and MIP models, enhanced with several novel acceleration mechanisms and Benders cuts. We provide analytical insights into the generalizability of our exact decomposition approaches for solving problems involving other coherent risk measures, as well as managerial insights. Our experimental studies reveal the following algorithmic and managerial insights: (i) monolithic MIP is effective only for small instances; (ii) the two-level decomposition and CP gracefully trade off solution accuracy with speed on small-to-medium instances; (iii) the three-level decomposition is the only approach that remains consistently scalable on large instances while still producing competitive incumbents and bounds.

Blood is an irreplaceable, life-saving resource that plays a pivotal role in healthcare systems worldwide. In Canada, nearly half of the population either requires blood or knows someone who does at some point in their lives. This study introduces the drone-aided mobile blood collection problem, which integrates mobile blood donation vehicles (bloodmobiles) with drones to improve operations related to the blood collection in urban areas. Each vehicle, carrying multiple drones, travels to several collection sites to conduct blood collection operations within a working day. Drones fly between vehicles to pick up collected blood bags and deliver them to the blood center. This collaborative framework enhances the performances of the collection system and ensures the freshness of collected blood upon arrival to the blood center. We develop a novel mixed-integer linear programming model for this problem that accounts for the age of each blood bag from the time it is donated to the time it arrives at the blood center. The proposed model optimizes the routing and scheduling of bloodmobiles and drones, along with blood collection and transportation operations over the course of a working day. The objective function maximizes the total collection reward, where the reward for each collected blood bag depends on its age when it reaches the blood center. To the best of our knowledge, this study is the first to consider this problem.
We also develop a hybrid matheuristic (HM) to solve large-scale instances of the problem. This algorithm combines a rolling horizon approach, which divides the problem into manageable subproblems solved sequentially, with a local branching technique that enhances solutions by exploring promising neighborhoods. To evaluate the algorithm’s performance, we conduct a comprehensive computational study. The numerical results reveal that Gurobi struggles to solve the model efficiently, yielding solutions with an average optimality gap of 47.83% across instances of varying sizes. In contrast, the HM algorithm delivers solutions with an average optimality gap of 4.02% that are, on average, 41.42% better than those of Gurobi. Notably, this enhancement is achieved alongside an average execution time reduction of 27,948.14 s. We also benchmark the proposed algorithm against the rolling horizon (RH), relax-and-fix (RF), and fix-and-optimize (FO) matheuristics. The results show an average improvement of 7.12% in the solutions achieved by the HM over the RH. When compared to RF and FO, the HM algorithm not only provides superior solution quality but also reduces execution times by an average of 6,413.54 s for instances with fewer than 30 collection sites. This computational efficiency becomes even more evident as the problem size increases, due to the inability of RF and FO to provide a feasible solution within the 10-hour time limit.
Finally, we demonstrate the real-life applicability of the problem through a case study in Quebec City, Canada. This case study examines the impact of fleet size on the performance of the blood collection system, compares operations with and without the integration of drones to highlight their benefits, and presents a sensitivity analysis on drone-related parameters. The results emphasize the importance of drones in improving the efficiency and reliability of the blood collection system. Most importantly, drones reduce the age of blood bags received at the blood center by shortening the time required for their delivery.

Degree-based topological indices in mathematical chemistry are numerical values used to analyze the structure of molecules and predict their physicochemical behaviour. Given any degree-based topological index, determining the chemical graphs which optimize that index reduces in determining the extreme points of a polytope. We focus on determining the chemical graphs of maximum degree at most 4 which optimize such indices.

The rapid growth of electric vehicles requires efficient charging management. Conventional optimization methods struggle to assign charging requests in real time due to high-dimensional system states. This project develops an intelligent reinforcement learning framework that dynamically assigns vehicles to charging stations while minimizing travel time, waiting time, and charging delays.

Key Words: Electric vehicle charging, Reinforcement learning, Real-time assignment

This work presents an Environment-Based Design (EBD)-guided hybrid LLM and prompt engineering pipeline for extracting requirements from medical device regulations. By integrating Recursive Object Model (ROM)-based prompts with heuristic normalization, the approach improves accuracy, consistency, and traceability. Preliminary results indicate improved accuracy and efficiency compared to manual and LLM-only methods, suggesting potential for scalable regulatory compliance.

Pharmaceutical supply chains face challenges in managing Unwanted Medicines (UMs), which pose economic, environmental, and regulatory risks. Coordinating multiple companies under Extended Producer Responsibility (EPR) involves linked decisions across pricing, waste collection, and emission reduction, with many variables to optimize simultaneously. Because each supply chain member’s decisions affect others and strategic interactions are important, a game-theoretic approach is required to capture these interdependencies. However, traditional game-theoretic methods, such as backward induction, cannot handle this level of complexity. This study proposes a hybrid framework that integrates game-theoretic modeling with a Gradient Ascent algorithm within a hierarchical game structure to solve equilibrium strategies when closed-form solutions are unavailable. The algorithm iteratively optimizes the manufacturer’s unit transfer prices across different medicine categories, which drive all dependent decisions. Starting from feasible values, the prices are updated until the algorithm reaches a stable solution, establishing a computational link between the leader’s decisions and downstream variables while enabling consistent optimization of pricing, collection incentives, decarbonization investments, and EPR compliance. A novel Hierarchical Conditional Benefit-Sharing (HCBS) contract is embedded and optimized within the framework, aligning actor incentives with sustainability targets. Coupled with Gradient Ascent optimization, it enhances UMs collection, promotes recycling and reuse, and improves overall closed-loop supply chain coordination. Finally, analysis of real-world data from the Canadian pharmaceutical supply chain for analgesics, antibiotics, and anticancer drugs demonstrates that the proposed circular strategies reduce medicine prices by 2% and lower emissions by 35.4%. Furthermore, the HCBS contract increases UMs collection by up to 42.4%, highlighting its effectiveness in improving collection rates and environmental performance. This work contributes a scalable optimization methodology for game-theoretic supply chain models, with broader applicability to closed-loop systems under regulatory complexity.

We propose a chance-constrained formulation for power system transmission expansion planning under uncertain generation and demand. The bilevel problem is solved via a novel graph optimization metaheuristic. A clustered Monte Carlo framework leverages Lagrange multipliers to screen scenario feasibility without solving full optimal power flows, enabling efficient and scalable estimation of operating costs and reliability penalties.

We study a Competitive Facility Location Problem with rank-based choice models to estimate customer behaviour. We reformulate the standard bilevel program, replacing primal assignment variables and constraints by their dual representation, to enable a Benders-inspired decomposition. Our branch-and-cut algorithm solves Quebec-inspired instances 10 times faster, on average, than baseline approaches.

Branch-and-Bound (B\&B) is the dominant exact solution method for Mixed Integer Linear Programs (MILP), yet its exponential time complexity poses significant challenges for large-scale instances. The growing capabilities of machine learning have spurred efforts to improve B\&B by learning data-driven branching policies. However, most existing approaches rely on Imitation Learning (IL), which tends to overfit to expert demonstrations and struggles to generalize to structurally diverse or unseen instances. In this work, we propose Tree-Gate Proximal Policy Optimization (TGPPO), a novel framework that employs Proximal Policy Optimization (PPO), a Reinforcement Learning (RL) algorithm, to train a branching policy aimed at improving generalization across heterogeneous MILP instances. Our approach builds on a parameterized state space representation that dynamically captures the evolving context of the search tree. Empirical evaluations show that TGPPO often outperforms existing learning-based policies in terms of reducing the number of nodes explored and improving p-Primal-Dual Integrals (PDI), particularly in out-of-distribution instances. These results highlight the potential of RL to develop robust and adaptable branching strategies for MILP solvers.