• A vessel-drone collaborative framework is proposed for multi-island port logistics. • Planning model incorporates island obstacles and port berthing constraints. • A CNN-driven Range Optimizer resolves drone range conflicts via en-route planning. • The method reduces total cost by 6.44%–7.98% and mission time by 16.45%–17.87%. The fragmented terrain and limited infrastructure in multi-island areas present significant engineering challenges for port logistics, including high operational costs, prolonged transport cycles, and excessive port congestion. To address these issues, this paper proposes a vessel-drone collaborative logistics system. Recognizing that existing vehicle routing models fail to account for island obstacles, we develop a mission planning framework grounded in Mixed-Integer Linear Programming (MILP) and a Convolutional Neural Network (CNN)-driven range optimizer. The framework involves: (1) modeling island terrain and port distribution with optional vessel routes via path planning; (2) formulating a MILP model for transport planning; and (3) applying a CNN-driven post-processing framework that uses constraint relaxation and en-route strategies to resolve conflicts between terrain obstacles and drone range limits. Numerical experiments validate the method’s effectiveness and provide planning examples for real-world scenarios. Compared with traditional vessel-only transport, the proposed approach reduces total system cost by 6.44%–7.98% and mission completion time by 16.45%–17.87% across varying island densities, with time savings of 17.65%–24.16% under different port berthing durations. These results underscore the engineering value of embedding ML-driven optimization into real-world port logistics systems, offering a scalable pathway toward intelligent and sustainable maritime operations in multi-island areas.
Sun et al. (Fri,) studied this question.