AI-Driven Collaborative optimization of Berth shifting, allocation, and vessel scheduling in restricted-channel ports for enhanced operational resilience

Abstract

Restricted-channel ports are grappling with critical operational issues due to the interaction of tidal restrictions, limited space, and increasing vessel flows. Ineffective berth allocation and vessel scheduling in these contexts trickle down to increased vessel waiting time, reduced throughput, and a lack of resilience to uncertainties. Here, we present AI-CBAVS—the AI-Driven Collaborative approach for joint Berth-shifting, Allocation and Vessel Scheduling—that leverages deep reinforcement learning (DRL), graph neural networks (GNN), and multi-agent reinforcement learning (MARL) in a real-time optimization framework. The GNN represents the topological information of the port as a dynamic graph, allowing the DRL agent to learn the optimal allocation policy under tidal, weather, and demand uncertainty. MARL resolves conflicts and orchestrates coordinated berth-shifting actions among diverse decision-makers - berth controllers, vessel traffic services, and pilot dispatch. The approach is evaluated in a digital-twin simulation of a 12-berth restricted-channel port, subject to 15 classes of disruption scenarios, over 1,000 Monte Carlo simulations. Benchmarking results show that AI-CBAVS cuts average vessel waiting time by 66%, boosts berth occupancy to 91.7%, increases daily vessel throughput by 72.5%, and reduces the average time to recover from disruptions from 142 to 31 minutes compared to traditional integer linear programming-based approaches. These results verify the feasibility and scalability of AI-CBAVS as a framework for next-generation smart ports.