A Multi-Resource Coordinated Scheduling Optimization Framework Integrating Berth Allocation, Quay Crane Operation, and AGV Transportation in Automated Container Terminals

This study investigates the operations in automated container terminals, in which scheduling decisions need to be generated automatically. Therefore, an integrated optimization model is established considering the two-way transportation process of containers between the berth and the yard with the assignment and scheduling of berths, quay cranes, and automated guided vehicles (AGVs). In particular, a buffer zone is added at the interface between the berth and the GV routing area. The mathematical model is divided into two stages, in which the first stage deals with the arrangement of vessels and quay cranes, while the second stage assigns container transport tasks to AGVs. Furthermore, a coordinated mechanism with an elite solution pool is introduced to coordinate the seaside scheduling and AGV transportation decisions. To solve the first stage, an adaptive large neighborhood search (ALNS) enhanced by reinforcement learning is proposed. The reinforcement learning strategy updates the selection rule for destroy-and-repair operator pairs. For the second stage, the AGV transportation environment is modeled as a directed graph, encoding the road network, the AGV interaction relationships, and the task states using a graph neural network (GNN). The AGVs are divided into several groups, mainly according to the initial locations, and each group serves nearby tasks. Therefore, multi-agent proximal policy optimization (MAPPO) is introduced with the GNN to arrange the AGV tasks. The proposed models and algorithms are evaluated through computational experiments with different problem scales, demonstrating better performance compared with the selected baseline methods under the tested settings.