Hierarchical planning and scheduling for bulk ports via network flow and deep reinforcement learning-guided constraint programming

In this research, an integrated inbound and outbound operational planning and scheduling problem is addressed for complex and large bulk ports. The practice of moving homogeneous dry bulk cargoes on a fixed terminal is changing as raw materials of different types are transported from/to the same terminals. It raises a new research challenge where unloading, stacking, reclaiming, conveying and loading operations must be coordinated to import/export blended products according to the tight specifications of customers. This paper aims to maximise resource utilisation and to satisfy demands as early as possible. The essence of the problem is to design the routing of product flows throughout the port logistics network such that supply and demand are matched optimally. This study presents a new framework that enables the modelling of the planning part as a multi-commodity flow problem and the scheduling part as a constraint programming (CP) problem. A novel dual-engine optimisation method that synergistically combines CP with deep reinforcement learning (DRL) is proposed to accelerate the scheduling phase. The method leverages DRL agents to fix key variables, thereby effectively accelerating the optimisation process of the CP solver. Comprehensive numerical experiments are conducted on real data sets as well as instances derived from real scenarios to validate the effectiveness of the proposed approach, demonstrating significant improvements in port scheduling efficiency. Additionally, strategic management analyses offer actionable insights to support decision-making in bulk port operations. The proposed methods provide a generalised methodology adaptable to a broad range of complex combinatorial optimisation problems in port logistics and beyond, paving the way for more intelligent and sustainable dry bulk port management.