Enhanced multi-agent approaches for efficient evacuation and rescue operations in managing disasters

• Main Author: Abusalama, Jawad
• Format: Thesis
• Language: English; English
• Published: 2025
• Subjects: Q Science; QA Mathematics
• Online Access: http://eprints.utem.edu.my/id/eprint/29012/

This study addresses disaster management within Multi-Agent System (MAS) environments, focusing on two critical phases: evacuation and rescue. The study tackles two primary challenges: the Emergency Route Planning (ERP) problem, which involves determining optimal evacuation routes within capacity-constrained transportation networks, and the Winner Determination Problem (WDP) in reverse combinatorial auctions, which pertains to effective task allocation and coordination among rescue agents. The research progresses through four stages: problem definition, approach design, implementation and evaluation, and simulation. For the evacuation phase, a Dynamic Real-Time Capacity Constrained Routing (DRTCCR) algorithm is introduced to address ERP challenges. The algorithm aims to generate optimal evacuation routes considering the complexity, capacity constraints, and scale of evacuees in the transportation network. Analytical evaluation against existing algorithms, specifically Multiple-Route Capacity Constrained Planner (MRCCP) and Max-Flow Rate Priority (MFRP), demonstrated that the DRTCCR significantly improves performance in terms of Total Evacuation Time (TET) and Weighted Average Time (WAT). Compared to MRCCP, DRTCCR reduced TET by 14.95% and WAT by 1.7%, while against MFRP, it decreased TET by 17.25% and WAT by 9.18%. In the rescue phase, two innovative approaches are proposed to enhance task allocation for WDP in reverse combinatorial auctions. These approaches were rigorously evaluated against Andrea’s algorithm and a Genetic Algorithm, revealing competitive advantages. Notably, as the number of bidders increased, the execution time of competing approaches escalated exponentially, while the proposed approaches exhibited a steady increase. Building on the proposed algorithm and approaches, Agent-Based Simulation (ABS) models were developed to evaluate both evacuation and rescue operations in Al-Aqsa Mosque (AM) scenarios in Palestine. The ABS evacuation model demonstrated superior performance compared to the Random, Kasereka, and Nearest Neighbor Search (NNS) models, achieving a 0% Total Deaths (TD) rate, outperforming Kasereka 1%, Random 5.5%, and NNS 14%. It also achieved a 99.5% Total Alive Evacuees (TA) rate, compared to 98.7% for Kasereka, 94.9% for Random, and 87.6% for NNS, along with an Average Health of Alive Agents (ATA) improvement of 52.4% over Kasereka, 82.1% over Random, and 93% over NNS. Similarly, the ABS rescue model outperformed both the Nearest Neighborhood Rescuing (NNR) model and the Hooshangi and Alesheikh model, reducing the duration of rescue operations by 49.2% compared to NNR and 32.6% compared to the Hooshangi and Alesheikh model, while also decreasing the number of casualties by 10.6% relative to NNR and 2.4% relative to the Hooshangi and Alesheikh model. These results highlight the model's significant improvements in both efficiency and effectiveness in managing evacuation and rescue scenarios.