Maritime transport is vital to the global economy, yet the frequency of natural disasters at sea continues to rise, resulting in more persons falling overboard. Therefore, effective maritime search and rescue (SAR) hinges on accurately predicting the probable distribution of drifting victims and on rapidly devising an optimal search plan. Conventional SAR operations either rely on rigid, pre-defined patterns or employ reinforcement-learning techniques that yield non-unique solutions and incur excessive computational time. To overcome these shortcomings, we propose an adaptive SAR framework that integrates three modules: (i) the AP98 maritime-drift model, (ii) Monte Carlo particle simulation, and (iii) a mixed-integer linear programming (MILP) model. First, Monte Carlo particles are propagated through the AP98 model to generate a probability density map of the victim’s location. Subsequently, the MILP model maximizes the cumulative probability of rescue success while minimizing a composite cost index, producing optimal UAV search trajectories solved via Gurobi. Experimental results on a 10 km × 10 km scenario with five UAVs show that, compared with traditional parallel-line search, the proposed MILP approach increases cumulative success probability by 12.4% within the first twelve search steps, eliminates path overlap entirely, and converges in 9.5 s with an optimality gap of 0.79%, thereby demonstrating both efficiency and real-time viability. When MIPFocus (a solver setting in Gurobi that controls the emphasis of the Mixed Integer Programming solver) aims at the optimal solution and uses the parallel solution method at the same time, the best result is achieved.
Zhang et al. (Tue,) studied this question.