In maritime search and rescue (SAR) operations, the estimated location of survivors spreads over time, causing the search area to expand continuously. Deploying limited search and rescue units (SRUs) efficiently is critical, but as the search area grows, the number of possible deployment combinations increases exponentially, making exhaustive search impractical. This study proposes a two-stage metaheuristic-based optimization framework that balances computational efficiency and solution quality. In the first stage, the search area is discretized into a grid using simulated particle diffusion results, and grid cell importance is estimated based on particle distribution to assign limited SRUs to high-priority cells first. In the second stage, SRU deployment within each selected cell is reformulated as a permutation-based matching problem to refine resource allocation. The proposed framework was evaluated across 72 scenarios using genetic algorithm, simulated annealing, particle swarm optimization, and differential evolution. Results show that genetic algorithm consistently outperformed other algorithms under resource-constrained conditions while maintaining high solution diversity, providing multiple high-quality alternatives for practical SAR decision-making.
Tae-Hoon Kim, H. Jeong, Choong-ki Kim et al.· Proceedings of the Genetic a...· 0 citations
Unmanned Aerial Vehicles play a pivotal role in maritime Search and Rescue (SAR) missions, yet generating optimal flight paths remains a significant challenge. Traditional meta-heuristic algorithms, such as genetic algorithms (GA), often suffer from premature convergence to local optima, while reinforcement learning (RL) approaches typically struggle with low sample efficiency and slow initial convergence. To address these limitations, this paper proposes a hybrid framework termed Memetic Actor-Critic (MAC). The MAC framework integrates the decision-making policy of Actor-Critic (AC) with the global search capabilities of a memetic algorithm. Specifically, we introduce a problem-specific local refinement mechanism that utilizes state-value estimations to refine the actor's policy, enabling precise exploitation of high-value regions beyond immediate rewards. The proposed method was thoroughly evaluated using a realistic SAR scenario constructed from oceanographic particle simulation data collected from the East Sea of South Korea. Experimental results demonstrate that MAC significantly outperforms standalone GA, MA, AC, and other variants in terms of both solution quality and convergence stability, proving its effectiveness in balancing global exploration and local exploitation.
Jingeun Kim, Seung-Yeol Hong, Yong-Hyuk Kim et al.· Proceedings of the Genetic a...· 0 citations