Container terminals play a crucial role in global logistics and trade, with gate-in operations significantly impacting overall terminal efficiency and cargo turnover speed. This paper analyzes a series of problems caused by the randomness of the arrival of export containers at the container yard, including wastage of yard space, excessive waiting time for external trucks, and conflicts with other production operations. To address these issues, a method based on a decomposed ensemble framework is proposed to predict short-term container quantities for gate-in operations at container terminal gates. The experiment compares the autoregressive integrated moving average (ARIMA) algorithm, the prophet algorithm, and the Long Short-Term Memory (LSTM) algorithm, with results indicating the clear advantage of Long Short-Term Memory in decomposed time series modeling. The introduction of this method is expected to enhance the accuracy and flexibility of terminal production planning, optimizing resource utilization. Contributions of this paper include the proposal of predictive models, a shipping route-based decomposed-ensemble framework, and confirmation of the superiority of Long Short-Term Memory in prediction through comparative analysis. These contributions are expected to improve terminal operational efficiency, reduce resource wastage, and better adapt to the highly stochastic gate-in operation environment.

With the expansion of automated container terminals (ACTs), joint scheduling among multiple types of equipment has become a critical factor affecting operational efficiency. This study investigates a joint scheduling optimization problem of quay cranes (QCs) and automated guided vehicles (AGVs) by considering AGV battery swapping strategies under spare battery constraints. With the objective of minimizing the final task completion time of AGVs, a mixed-integer programming model is formulated that simultaneously accounts for task assignment, operation sequencing, battery swapping thresholds, spare battery quantity, and mutual waiting times between AGVs and QCs. To solve this problem efficiently, a hill-climbing genetic algorithm (HC-GA) is proposed. Numerical experiments under different task scales show that HC-GA outperforms the genetic algorithm (GA), simulated annealing (SA), Q-learning, and the Q-learning-based genetic algorithm (Q-GA) in key indicators. In addition, the experimental results show that a proper configuration of AGVs can improve scheduling coordination and enhance the energy utilization efficiency of AGVs. The number of spare batteries and the threshold have significant impacts on overall system performance. When both operational efficiency and equipment utilization are considered, appropriately configuring the number of spare batteries and the threshold can effectively enhance the operational efficiency of ACTs.