4 Presentations

• 
  10:30 AM - 10:55 AM

  Real-time data-driven inventory management in a hospital pharmacy

  • Ali Jafari, presenter, Polytechnique Montreal
  • Antoine Legrain, Polytechnique Montréal
  • Nadia Lahrichi, Polytechnique Montréal

  The provision of high-quality and cost-effective healthcare is a major global issue, and managing medication supply chains in hospital pharmacies is a complex task due to factors such as space limitations, demand uncertainty, and medication shelf-life. This research addresses the pharmacy inventory control problem at Centre Hospitalier de l'Université de Montréal (CHUM), Canada. This research aims to enhance the efficiency of medication inventory management within the hospital by optimizing the inventory management policy and choosing the best inventory control parameters for each group of medicines in real-time. The project involved analyzing demand and categorizing items in the care unit into fast and slow-moving products. A novel real-time data-driven inventory policy is proposed for medications at the point-of-use location in the hospital, considering demand uncertainty and space constraints. The proposed stochastic optimization model is based on the (s,S) policy, with fast-moving drugs continuously reviewed and the only ones that can initiate the CU's replenishment process, while slow-moving drugs are only reviewed when a fast-moving item needs replenishment. To solve the large stochastic problem, the model is decomposed based on the categories derived from the data, and a rolling-horizon method is implemented to capture demand fluctuations and update inventory control parameters. 
• 
  10:55 AM - 11:20 AM

  A Prediction-Based Approach for Online Dynamic Appointment Scheduling: A Case Study in Radiotherapy Treatment

  • Tu-San Pham, presenter, Polytechnique Montreal
  • Antoine Legrain, Polytechnique Montréal
  • Patrick De Causmaecker, Katholieke Universiteit Leuven Campus Kortrijk
  • Louis-Martin Rousseau, Polytechnique Montréal

  Patient scheduling is a difficult task involving stochastic factors, such as the unknown arrival times of patients. Similarly, the scheduling of radiotherapy for cancer treatments needs to handle patients with different urgency levels when allocating resources. High-priority patients may arrive at any time, and there must be resources available to
  accommodate them. A common solution is to reserve a flat percentage of treatment capacity for emergency patients. However, this solution can result in overdue treatments for urgent patients, a failure to fully exploit treatment capacity, and delayed treatments for low-priority patients. This problem is especially severe in large and crowded hospitals. In this paper, we propose a prediction-based approach for online dynamic radiotherapy scheduling that dynamically adapts the present scheduling decision based on each incoming patient and the current allocation of resources. Our approach is based on a regression model trained to recognize the links between patients’ arrival patterns and their ideal waiting time in optimal off-line solutions when all future arrivals are known in advance. When our prediction-based approach is compared with flat-reservation policies, it does a better job of preventing overdue treatments for emergency patients and also maintains comparable waiting times for the other patients. We also demonstrate how our proposed approach supports explainability and interpretability in scheduling decisions using Shapley additive explanation values.
• 
  11:20 AM - 11:45 AM

  An exact linearization of mixed-binary quadratic model predictive control for real-time demand response

  • Fatemeh Rajabi, presenter, Polytechnique Montreal
  • Antoine Legrain, Polytechnique Montréal
  • Antoine Lesage-Landry, Polytechnique Montréal

  Frequency regulation is a fast-timescale power balancing service which is growing in importance given the increasing penetration of renewable in power systems. Demand response can be utilized for frequency regulation through the control of thermostatic controlled loads. For instance, a heating, ventilation, and air conditioning (HVAC) system can be sequentially turned ON and OFF to modulate the load’s consumption. In this work, we consider a large aggregation of loads, and aim to track a setpoint representing a power imbalance, e.g., due to renewable intermittenary, while always keeping the loads temperature within acceptable bounds. We consider the lockout constraint which prevents successively turning ON and OFF an HVAC. We formulate the demand response as a model predictive control problem to track the regulation setpoint while accounting for the load dynamics. The MPC problem takes the form of a mixed-binary, multi-period quadratic program which is ill-suited to fast-timescale decision making. We propose an exact linearization to reformulate the problem into a mixed-binary linear program. This results in a significant reduction in running time as well as a remarkable computational resource deduction without compromising the quality of the solution. Preliminary results generated from the simulation will also be presented.
• 
  11:45 AM - 12:10 PM

  Offline Solution based policies for on-call personnel scheduling

  • Prakash Gawas, presenter, Polytechnique Montreal
  • Antoine Legrain, Polytechnique Montréal
  • Louis-Martin Rousseau, Polytechnique Montréal

  About 1.5% of workers in North America who are involved in on-call employment suffer from issues like irregular hours, and lack of flexibility. We present a novel on-call personnel scheduling system and aim to optimize its operation with a dynamic, data-driven approach to alleviate some of these issues. Like traditional systems, it contacts on-call personnel in order of seniority to inform them about the availability of on-call work. However, flexibility in the system allows employees time to think. It also allows senior employees to replace or bump junior employees from their preferred shift if a junior employee had responded early and selected that shift. The replacements are undesirable and the management seeks a policy to make calls to employees to schedule all available shifts while ensuring minimum bumps under uncertain employee response times. We show that this problem is NP-complete even in the case of perfect information. Then, we design easy-to-implement policy approximations based on a threshold structure for the dynamic problem and tune them using offline solutions obtained from simulated data, where all uncertainty is known beforehand. The offline solution-based policies are tested on real-world data and outperform the current policy used by our industrial partner.