Predict-then-optimize for automated guided vehicle opportunity recharging scheduling considering nonlinear charging time

Automated container terminals deploy electric automated guided vehicles (AGVs) for horizontal transportation. Effective AGV fleet management requires precise scheduling of recharging operations. This process is complicated by the imprecision of charging times and the limited charging piles. Oversimplified linear assumptions and conventional threshold-based charging strategies often lead to inaccurate charging time estimates and sub-optimal operational policies, respectively. This paper addresses the AGV opportunity recharging scheduling problem with nonlinear charging and limited charging piles by developing a predict-then-optimize framework. A mixed-integer nonlinear programming model is developed to minimize the total task delay while ensuring battery safety and respecting charging pile capacity limits. A Transformer-based predictor is used to evaluate nonlinear charging durations from historical data within the scheduling procedure. To solve the model, a matheuristic combining adaptive large neighborhood search and local search is proposed. This algorithm operates through a coordinated three-layer structure. An adaptive large neighborhood search handles the global exploration of charging decisions. A three-stage mixed-integer linear programming method conducts the exact evaluation of charging timings and service sequences. Finally, the local search further optimizes the schedules. Computational experiments using real terminal data demonstrate that the Transformer model achieves high predictive accuracy with an R 2 value of approximately 0.99. The proposed framework demonstrates high computational efficiency across practical scales and outperforms conventional linear charging models by mitigating the delays induced by simplified approximations. These results provide actionable insights for scheduling AGV charging operations across varying operational scales.