3 Presentations

• 
  03:30 PM - 03:55 PM

  Developing a Robust Multi-objective Optimization Model for Reverse Logistics of Battery Electric Vehicles

  • Aysan Mahboubi, presenter, Polytechnique Montreal
  • Roozbeh Yaghoubi, Polytechnique Montreal
  • Mina Kazemi Miyangaskary, Polytechnique Montreal
  • Samira Keivanpour, Polytechnique Montréal

  In response to the damaging impacts of greenhouse gas emissions, Electric Vehicles (EVs) have emerged as a sustainable alternative. However, the rise of EVs creates the challenge of disposing numerous retired batteries. Hence, developing a robust multi-objective optimization model for Reverse Logistics (RL) of Battery Electric Vehicles (BEVs) can be an effective approach. Existing uncertainties, however, cause a gap between the outputs of exact optimization models and real-world conditions. Addressing this, the present study develops a robust multi-objective optimization model that incorporates uncertainty, making it more applicable to real-world scenarios. This research compares the relative performances of deterministic and the proposed robust optimization models and validates the model by calculating violation probabilities within a robust optimization framework (budget of robustness). Additionally, this study explores adjusting the level of conservatism in decision-making by applying the price of robustness approach to manage conservatism in our decision-making process. Highlighting the importance of innovative sustainable practices, this work offers a practical route for stakeholders to collaboratively mitigate the challenges associated with the end-of-life management of EV batteries, thus contributing to environmental sustainability and resource efficiency in the BEV industry.

• 
  03:55 PM - 04:20 PM

  Distributionally Robust Warm-Starting for Mineral Supply/Value Chains

  • Yassine Yaakoubi, presenter, McGill University
  • Roussos Dimitrakopoulos, COSMO Stochastic Mine Planning Laboratory, Université McGill

  Grounded in mining applications, this seminar presents a distributionally robust warm-starting mechanism to streamline optimization. Using historical solutions, a predictive model is trained, deploying graphical models to learn (mining) block-based connectivities. Subsequently, resource-constrained min-cost max-flow problems are solved to formulate initial production schedules, thus capturing the structure (spatial distribution) and uncertainty inherent in mining operations, and more specifically in the simultaneous stochastic optimization of mining complexes (mineral supply/value chains). The outcomes are twofold: a set of distinct solutions, each reflecting a specific demand scenario, and/or a singular risk-aware solution informed by block connectivity probabilities. This approach shifts warm-starting from a deterministic single point estimate, the norm in the mining literature, to a distributional perspective. It demonstrates significant computational efficiencies and promises broad application in various domains for robust decision making under uncertainty. Finally, we present insights, potential limitations, and future research directions for developing robust, reasoning, and responsible decision support systems for industrial-scale uncertain environments.

• 
  04:20 PM - 04:45 PM

  A Risk-aware Location-Allocation-Pricing Problem with Stochastic Price-Sensitive Demands

  • Maryam Daryalal, presenter, HEC Montréal

  We consider a capacitated location-allocation-pricing problem in a single-commodity supply chain with stochastic price-sensitive demands, where the location, allocation and pricing decisions are made simultaneously. Under a general risk measure representing an arbitrary risk tolerance policy, the problem is modeled as a two-stage stochastic integer program (2SIP). To solve the problem, we adapt an integer L-shaped method for 2SIPs with a translation invariant monotone risk measure and demonstrate its applicability on three common risk measures for moderate, cautious and pessimistic policies. We also introduce two family of problem-specific cuts to enhance the performance of the algorithm. Our computational experiments are conducted to study the effectiveness of the algorithm, along with a comparison between the impact of different policies on the solution of the problem. Our approach for incorporating the risk and the proposed solution framework is general and can be used for other similar discrete optimization problems that make use of risk measures.