±«Óãtv

1300-1310

1310-1335

1335-1400

1400-1425

Ìý

Abstracts

Ìý

Ìý

Maximum Expected Time to Rescue – Helicopter Transit (METR-HT) in the Canadian Arctic

Isabella Fernandes, MASc. student

The arctic sea ice is shrinking at higher rates every decade, a consequence of well-documented climate change rates. Consequently, shipping activities in polar regions have increased, which has introduced new challenges for Search and Rescue (SAR) operations. This work proposes a simulation model to calculate the Maximum Expected Time to Rescue – Helicopter Transit (METR-HT) in the Canadian Arctic, a geographically vast and environmentally harsh region. The study focuses on selected scenarios for a range of weather conditions, number of people in distress, and incident locations. To achieve this, extensive interviews with subject matter experts were performed to develop a realistic simulation model and results. The simulation model considers various factors, including travel times, air bases, refueling stops, and weather-related thresholds for helicopter operationality. In addition, historical weather data has been analyzed to provide a firm empirical basis for the helicopters’ operability in the simulation model for the different scenarios. The resulting information on METR-HT can be used to enhance emergency preparedness, for instance to inform decisions on minimum survival supplies required for those in distress. Providing a platform to simulate the air response for different scenarios, the model can furthermore be used in follow-up research, for instance to study the effect of adding additional helicopters, bases, or fuel caches. This will further support the sustainable development of Arctic communities, for whom effective SAR is essential to support their land-based and marine activities.

Ìý

Assessing completion time of on-scene rescue missions in the Canadian Arctic

Parsa Rezaei, MASc. student

The objective of this research is to create a unique model specifically designed to estimate the duration of rescue missions in maritime incidents in the Canadian Arctic. The primary focus is on the time taken from the arrival of the rescuers at the incident scene to the successful completion of the rescue operation. The analysis will consider various factors influencing the duration of these missions, such as environmental conditions, different tasks involved, equipment utilized, types of responders participating, and the condition of the individuals in distress. This model will extensively rely on the expertise and experiences of professionals to comprehend and incorporate these intricate details. It employs a combination of discrete event simulation models, effectively integrating risk analysis, decision analysis, and data analysis. The model aims to provide estimates concerning the duration of rescue missions in the Canadian Arctic, addressing a notable absence in existing research.

Ìý

Reinforcement learning-based inventory & transshipment planning in a Physical Internet supply chain

Simranjeet Singh Chadha, PhD student

This study addresses the inventory replenishment and transshipment problem between two hubs within the Physical Internet (PI) paradigm using a Reinforcement Learning (RL) approach. The PI paradigm aims to improve logistics efficiency through open, interconnected, and sustainable networks. Traditional optimization methods, such as linear programming, often fall short in handling the dynamic and complex nature of such systems. We propose an AI-driven solution leveraging Q-learning, a type of RL, to minimize total costs, including replenishment, transshipment, and holding costs, while aligning with PI principles. The environment simulates two hubs with initial inventory levels, demands, and associated costs. The RL agent learns optimal replenishment and transshipment actions to balance inventories and meet demands efficiently within an open and dynamic logistics network. The state space comprises inventory levels at both hubs, while the action space includes possible replenishment and transshipment decisions. The reward function inversely correlates with the total costs, incentivizing cost minimization. The RL agent demonstrated significant improvements in cost efficiency. Performance was evaluated via total rewards and inventory management during a test phase. Our findings showcase the potential of RL within the PI paradigm, offering a robust alternative to conventional logistics optimization methods. This study underscores the feasibility of AI-driven strategies to enhance logistics efficiency and sustainability in the interconnected PI framework. Future work could explore advanced RL algorithms and real-time adaptations for further improvements in dynamic PI environments.