»ÆÉ«Ö±²¥

1100-1125

1125-1150

1150-1205

Break

1205-1230

1230-1255

Qixuan Zhao
Assessing the Suitability of Traditional Machine Learning Metrics for EMS Simulation Development

Ìý

Abstracts

Ìý

Ìý

A Bi-objective Approach for Hydrogen Refueling Station Optimization under Uncertainty

Mina Valaei, MASc Student

The development of hydrogen refueling infrastructure is essential for promoting the adoption of hydrogen vehicles. This study presents a bi-objective optimization model for the location-allocation of hydrogen refueling stations, considering minimizing costs while maximizing driver preference. The model incorporates demand prediction based on demographic factors influencing the adoption rate of hydrogen vehicles. To address uncertainties in drivers' travel patterns, a robust optimization framework is applied using the phi-divergence method. Due to the high computational time required to solve the robust problem, Benders Decomposition method has been implemented as the solution approach. Additionally, due to fire hazards associated with hydrogen refueling stations, the Technique for Order of Preference by Similarity to Ideal Solution (TOPSIS) ranking method has been used to assess and mitigate zonal risk. A case study in Halifax, Nova Scotia, demonstrates the model's applicability, balancing infrastructure development with predicted demand and considering driver behavior uncertainties. The results emphasize the need to account for travel pattern variability and demographic influences in optimizing refueling station locations, ensuring economic efficiency and user satisfaction.

________________________________________________________________________________

Improving Interorganizational Risk Management in Canadian Port Operations: A STAMP-Based Control Structure and Taxonomy

Elvira Melendez, PhD Candidate

This research explores the development of a framework to enhance interorganizational risk management (IRM) in Canadian port operations, addressing limitations on Systems-Theoretic Accident Model and Processes (STAMP) model. Utilizing STAMP, the study aims to construct a Port Risk Control Structure (PRCS) in Objective 1, forming the foundation for subsequent analyses. Building on this foundation, Objective 2 takes an exploratory approach to identify key IRM challenges through a targeted literature review and stakeholder interviews. The insights gathered inform Objective 3, which focuses on creating a tailored taxonomy, designed to systematically address PRCS limitations by categorizing IRM challenges and establishing protocols for enhanced interorganizational coordination. The research methodology includes validation through case studies and semi-structured interviews with Canadian port stakeholders and subject matter experts. Objective 4 assesses the taxonomy's practical relevance and usability, incorporating feedback to refine its application for broader use across diverse port environments. The taxonomy aims to strengthen the PRCS by streamlining safety processes, promoting continuous improvement, and aligning organizational safety practices with the realities of complex, multi-stakeholder operations. Finally, this research contributes to a holistic approach in IRM for port operations, bridging theoretical models and practical challenges to foster proactive risk management. By integrating Lean Six Sigma principles, it also emphasizes operational efficiency, making the taxonomy adaptable to the evolving demands of port safety and risk management practices.

________________________________________________________________________________

Capacity Planning and Patient Assignment in Long-Term Care Facilities: A Multi-Period Mathematical Model with Interfacility Transfers

Mahsa Pahlevani, PhD Candidate

Capacity planning in long-term care (LTC) facilities is essential for increasing healthcare system efficiency and addressing hospital overcrowding. Having sufficient LTC capacity, healthcare systems can improve patient transitions from hospitals to appropriate care facilities, reducing the burden on hospital resources and enhancing overall patient flow. This study presents a mathematical model for capacity planning in LTC facilities over a multi-period planning horizon. The objective is to determine optimal capacity expansions and patient assignments while balancing costs associated with expansion, waitlists, and unmet demand. Patients, categorized into two groups from both communities and hospitals have specific LTC preferences before assignment. The model accounts for expansion costs and allocates patients to preferred facilities. However, due to high expansion costs and limited capacity, patients may be placed on waitlists or assigned to non-preferred facilities, both incurring penalties. The problem is presented using two models: first, a base model without interfacility transfers is developed, having promising results in capacity planning and patient assignment. For the more complex model that includes interfacility transfers, a column generation approach is used to manage the increased problem size. Analysis shows that this enhanced model can identify optimal expansion planning for LTC facilities, improve patient assignments, and support efficient interfacility transfers.

________________________________________________________________________________

Assessing the Suitability of Traditional Machine Learning Metrics for EMS Simulation Development

Qixuan Zhao, PhD Candidate

Emergency medical services (EMS) is a critical component of the healthcare system. Simulation is a common approach to study the EMS system. Machine learning has gained success across various fields, leading to increased interest in combining simulation with machine learning-based input parameters to enhance simulation’s credibility. Traditionally, machine learning models used for EMS simulation input parameters are evaluated with standard performance metrics, such as accuracy and F1-score for classification tasks and mean absolute error for regression tasks. However, it remains unclear whether using traditional performance metrics for machine learning help simulation reflect the real-world EMS system in addition to other considerations. This study addresses this gap by assessing the suitability of traditional machine learning performance metrics for EMS simulation development through a real-world case study based in Nova Scotia, Canada.

________________________________________________________________________________

Ìý