Closed-loop receding-horizon search planning for USV-based maritime moving-target search under time-varying ocean currents

Maritime moving-target search under ocean-current disturbances is challenging because the target state is partially observable and the target probability distribution evolves with environmental drift and sensing feedback. To address this problem, this paper develops a closed-loop receding-horizon search framework for unmanned surface vehicle (USV) search planning in dynamic ocean environments. The framework integrates current-driven target propagation, negative-observation-based belief updating, probability-of-containment (POC) representation, and constrained trajectory optimization. In each replanning window, the posterior belief is updated using the executed search segment, and a belief-guided trajectory optimization problem is formulated by balancing cumulative detection probability and normalized travel cost under kinematic and short-horizon progress constraints. To solve the resulting time-varying and multimodal optimization problem, an improved Crested Porcupine Optimizer, termed LGMS-CPO, is proposed. It incorporates hybrid initialization, sine–cosine-enhanced local exploitation, information-geometric Gaussian updating, and multi-subpopulation elite migration to improve search coverage, convergence behavior, and population diversity preservation. Six benchmark scenarios are constructed using real ocean-current data, covering unimodal, bimodal, and trimodal target priors. Comparative experiments show that LGMS-CPO achieves higher objective values, generates more effective search trajectories, and exhibits favorable first-window convergence performance. Additional robustness evaluations under target maneuverability and obstacle constraints, together with ablation studies, further support the proposed framework.