【作者】 曾庆成；

【导师】 杨忠振；

【作者基本信息】 大连海事大学 ， 交通运输规划与管理， 2008， 博士

【摘要】 集装箱码头吞吐量的迅速增加增加了其作业调度的复杂性,因此需要相应的调度模型与方法来辅助支持集装箱码头作业的优化调度。目前尽管诸多学者就集装箱码头作业调度的展开了大量的研究现,但是如何有效处理各种约束条件和复杂的相互关系,如何提高模型求解效率,如何实现作业系统各个环节之间的协同调度,如何集成各种调度方法与技术等问题一直没有得到较好的解决。针对目前研究中的问题,本文从两个方面开展研究,一方面是集成调度模型的建立,力争通过集成调度模型提高作业系统各环节的协调性,实现装船与卸船作业的协同调度,从而提高作业效率;另一方面是调度方法的集成,拟通过对优化算法、仿真技术、强化学习等方法的集成,提高对复杂系统的描述能力,提高模型的求解效率。本论文首先从装卸作业流程的整体出发,把两个以上的决策复合在同一模型中,建立堆存位置-集卡调度优化模型,并设计求解调度方案的两阶段禁忌搜索算法。在此基础上,建立装卸作业序列优化的集成调度模型,同时优化装卸桥、集卡、与龙门吊的作业序列,并设计求解模型的混合优化算法。其次,建立装卸混合作业调度模型,通过装货船舶与卸货船舶的协同调度,实现集卡在不同“作业线”间的共享,从而减少集卡的空驶,提高集卡的利用率;同时设计基于两阶段禁忌搜索的算法。然后,研究装卸桥在同一贝位内同时进行装船与卸船作业的调度问题,即同贝同步装卸调度问题。分别以出口集装箱积载计划是否确定为假设条件,建立同贝同步装卸调度模型,并设计基于双层遗传算法的求解算法。接下来,集成优化算法的智能决策机制与仿真模型的评价功能,提出集装箱码头作业调度仿真优化法。为提高仿真优化的计算效率,设计基于遗传算法与神经网络的混合优化算法。同时将仿真优化用于装卸作业中的设备作业序列优化,以及同贝同步装卸调度中装卸桥作业序列与集装箱积载计划的优化。最后,将强化学习方法应用于集装箱码头作业调度,并将其与仿真相结合,通过仿真模型构建作业调度系统环境,采用Q-学习算法(强化学习方法的一种)和仿真模型交互获得最优调度方案。更多还原

【Abstract】 With the rapid increase of container terminal output in China, the operation scheduling becomes more and more complex. Thus scheduling models and algorithms are needed to support the operation practice. Issues related to container terminal operations have gained attention and have been extensively studied recently due to the increased importance of marine transport systems. However, how to tackle the complex constraints and interrelation; how to improve the computation efficiency; how to realize coordination of different sub-process; and how to integrate different scheduling methods and techniques, are the problems that have not been solved well.Considering the problems in existing studies, we will study the operation scheduling problem from two aspects. The first aspect is model integration; we improve the coordination of different sub-process and realize the integration of loading and unloading operations by integrating scheduling model, and thus improve the operation efficiency of container terminals. The second aspect is method integration; by the integration of optimization algorithm, simulation technique, and reinforcement learning, the description capability to complex system can be enhanced, and also the model solution efficiency can be improved.Firstly, from the aspect of whole operation process, we consider two sub-decisions in one model. An optimization model considering storage location and yard trailer scheduling is developed, and a two-phase tabu search algorithm is designed to solve the model. Based on this, an integrating scheduling model is developed to optimize the operation sequence of quay cranes, yard trailers and yard cranes simultaneously. And a hybrid algorithm is designed to solve the model.Secondly, a "multi-crane oriented" scheduling model to coordinate the operation of loading ship and unloading ship is developed. By this method, the yard trailers can be shared by different quay cranes. Therefore, empty drive of yard trailers can be decreased and the utilization ratio of yard trailers can be improved. Meanwhile, a two-phase tabu search algorithm was designed to solve the model.Thirdly, we study the problem that quay cranes perform loading and unloading operations simultaneously in the same ship-bay, namely quay crane dual cycling scheduling problem. Two models for quay crane dual cycling scheduling problem are developed, namely scheduling model supposing the loading plans are given; and scheduling model considering optimization of loading plans. To solve the models, a heuristic method, called bi-level genetic algorithms is designed.Fourthly, integrating the intelligent decision mechanism of optimization algorithm and evaluation function of simulation model, a simulation optimization method for operation scheduling in container terminals is proposed. To improve the computation efficiency, a hybrid optimization algorithm based on genetic algorithm and neutral network is designed. Meanwhile, simulation optimization is used to optimize the operation sequence of different equipments in loading and unloading process, the quay crane sequence and loading plan of outbound containers in quay crane dual cycling.Lastly, reinforcement learning is used to operation scheduling in container terminals, and also it is integrated with simulation technique. In this method, simulation model is used to construct the system environment, and optimal scheduling scheme is obtained by the interaction of Q-learning algorithm and simulation environment.更多还原