Team management, particularly in critical situations, is one of the most challenging aspects of relief and rescue efforts. In such contexts, rapid decision-making, efficient team coordination, and optimal allocation of activities and responsibilities are vital to mission success and minimizing casualties. It is therefore crucial to identify and apply scientific and technological methods to enhance the performance of rescue teams. The purpose of this study is to investigate and improve the operational performance of relief and rescue teams in post-earthquake scenarios. Because of the widespread damage they cause, the geographic extent of the affected areas, and the severe time constraints involved in rescuing victims, earthquakes are among the most complex and critical natural disasters. In these conditions, any improvement in operational management and decision-making can increase the speed of response actions and ultimately save more lives.
In this study, by integrating modern spatial sciences with artificial intelligence algorithms, a performance optimization model for rescue teams was developed. Ant Colony Optimization (ACO) was identified as one of the most effective collective intelligence methods for addressing complex, team-based decision problems. Reinforcement learning, specifically using the Q-Learning algorithm, was incorporated into the ACO framework to further improve its adaptability to the environment. The proposed model was enhanced by this hybridization to enhance task allocation and decision-making.
To implement and evaluate the model, key parameters such as the structure of the team, the expertise of its members, the distribution of tasks, as well as the extent of human and structural damage were identified. An earthquake response scenario was simulated in District 4 of Tehran's Region 3 with 70 trained rescuers in 14 five-member teams following Iranian Red Crescent protocols. A primary input to the model was data on location, damage intensity, team specialization, and search-and-rescue activities. Based on movement speed, task duration, distance to operational sites, and overall rescue success rate, reward and cost functions were developed.
Comparing the hybrid model with the standard ACO algorithm, the proposed hybrid model increased overall rescue efficiency by 19%. The findings of this study demonstrate that integrating collective intelligence with reinforcement learning offers a promising framework for managing rescue teams intelligently and adaptively during critical events. It can also be applied to other natural or human-induced disasters, providing a pathway for the development of intelligent decision-support systems for crisis management at local and national levels. |
