【作者】 江建宇；

【导师】 王爱虎；

【作者基本信息】 华南理工大学 ， 物流工程与管理， 2014， 博士

【摘要】 本文提出了一个针对共享腹地港口群集疏运系统的物流仿真模型，由三部分组成：第一，通过对比集疏运系统相关问题的传统解法优缺，阐释从智能体仿真建模切入之依据；第二，基于国际物流实践与供应链流程，研究集疏运系统的结构特征、局限条件与智能体行为模式，进而构建集成仿真平台；第三，以广东省东莞市的出口集装箱物流即“集运配流”为实验，检验所造模型，并指出其具体应用框架与实践意义。为证明基于Agent仿真研究共享腹地港口群集疏运系统之必要性，首先探讨了共享腹地港口群的研究概况，发现学界极少从微观的集疏运视角进行深入探讨；进而回顾了集疏运系统的有关文献，指出当前问题主要为系统构成理解狭隘、定量研究稀缺、方法缺乏充分依据，等等。为此，分别从数学建模与系统仿真切入，借鉴了物流、供应链、交通运输等相关领域分析，系统归纳出解析模型的各种方法缺陷，及仿真建模在内容上的缺失，并指出集疏运系统为典型的CAS复杂系统结构。鉴于从复杂自适应性视角对集疏运系统建模属当前的研究空白，接着围绕二次调研数据和系统论、国际物流理论，对港口群腹地集疏运的系统结构进行了详细分析。参考进出口供应链流程，归纳出自工厂订单接收到码头装船离港的8个集运出口环节；基于多主体决策视角，提出了共享腹地港口群集运系统之CAS总体框架，以及港口、国际海运、公路、水路、铁路集疏运、进出口监管、货主的7子系统组成结构；进而从主体智能行为与局限条件切入，指出了各子系统的核心Agent、决策内容、影响关系，与时间、费用等方面的重要约束，包括：选港订舱、集运方式、做柜时间、集运路线4大决策；工厂、港口、船公司间的直接/间接服务关联；通关决策流程；驳船/拖卡/海铁联运3类集运成本的估算公式；乃至拖车固定成本、铁运“一口价”、海运THC、THCS、O/F、Local Charges和查柜成本、班轮船期、驳船航线、专列班次等实践经验值，为仿真建模与数据实验提供基础。MAS建模阶段，围绕共享腹地港口群集疏运系统的CAS结构，提出了以多智能体仿真为核心，由ASM智能体生成、BA智能行为算法、SRM仿真系统运行、IM/OM输入、输出与SSCM仿真控制管理6子模块组成的ISPHT开放集成平台/模型框架，继而描述了集运系统仿真建模中“核心类”与“约束类”智能体的主从关联，Multi-agent模型与集成平台的关系，及ISPHT各模块的重点设计内容。在此基础上，针对港口群腹地集运智能体系统中的选港订舱和集运方式、时间、路线选择共4项决策，分别构建了具体算法，包括：受交货方式、港口船期与“做柜剩余时间”、集运出口“综合成本”、“顺利程度”与主体行为“惯性”等因素约束的选港决策模型selectPort()，藉机器学习理论开发的智能选港Q算法qLearningPort()；参考集疏运系统成本结构、估算公式的公路集运roadTransPricing()、水运waterTransPricing()、铁路集运railTransPricing()报价模块，和集运方式决策模型selectTransMode()；借鉴集运出口流程的甘特图分析的做柜时间决策模型selectVanningTime()；以及由集运公路网分层设计算法、Floyd算法改造的静态集运线路决策模型selectStaticTransPath()，与以Dijkstra算法为核心的实时路由更新模型rtTransRouteUpd()。各模型与算法大部分基于“线性”函数与规则构建，有效弥补了解析模型建模难、物流局限考虑不全、求解复杂度高等缺陷，体现了从微观Agent智能体仿真视角研究集疏运系统的创新意义。为检验模型，实验阶段，通过程序代码实现了ISPHT平台的公路集运模块，在此基础上，以东莞市腹地港口群为对象，展开计算实验。实验结果表明，所构造模型可方便观测各主干道路与具体路段的实时、累计集装箱流量分配值，并与现实情况相吻合，证实了本文所设计的共享腹地港口群集疏运系统MAS仿真模型的合理性，以及基于智能体仿真方法分析、优化集疏运系统的可行性与实践价值。更多还原

【Abstract】 This paper presents an agent-based model for freight transportation systems of contestablehinterlands among a cluster of ports, which is composed of three parts: At first, by comparingthe advantages and disadvantages of the traditional solution of collection and distributionsystem related issues, to explain why using the agent based modeling and simulation;secondly, based on the practice of international logistics and supply chain processes, researchcollection and distribution system structural characteristics, conditions and limitations ofagent behavior, and then build an integrated simulation platform; thirdly, make experiment onDongguan hinterland to test the model, and pointed out their optimization framework.In the literature review chapter, the paper first discusses the hinterland of the port groupshared Research Survey found that scholars rarely discuss in depth from collecting anddistributing micro perspective; further review of the relevant literature collection anddistribution system, pointing out that the current system constitutes a major problemunderstanding narrow, the scarcity of quantitative research methods lack sufficient basis, andso on. To this end, the researchers were from the mathematical modeling and systemsimulation cut, drawing related fields of logistics, supply chain and transportation analysis,system deficiencies summarized various methods analytical model, and simulation modelingmissing in content, and pointed out that the collection and distribution system for a typicalCAS complex system architecture, proved the necessity and innovative Agent-basedsimulation of the hinterland of the port cluster sharing and distribution systems.Then around the secondary research data and systems theory, international logistics theory,the collection and distribution system structure is analyzed in detail. Referring first to importand export supply chain processes, factory orders received by inductive loading dockdeparting8Cargo export sector; followed by multi-agent-based decision-making perspective,the general framework proposed CAS cluster shared port hinterland transport systems, as wellas ports, international sea, road, waterway, railway transportation, import and exportregulation, the owner of seven subsystems structure; thus cut from the body of the conditionsand limitations of intelligent behavior, pointed out the various subsystems of the core Agent,decision-making content, affect the relationship with time important constraint costs and otheraspects, including: election harbor booking, Cargo way to do cabinets time, Cargo route4bigdecision; directly or indirectly related factories, ports, shipping companies inter/services;clearance decision-making process; barge/trailer/sea-rail intermodal transportation costs threecategories set estimation formula; fixed costs and even trailers, railway transport,"one price", sea THC, THCS, O/F, Local Charges and counter check costs, liner sailing barge routes, trainfrequency, etc. practical experience, for the following simulation modeling and experimentaldata provide a solid foundation.MAS modeling stage, clustered around a shared hinterland of the port of CAS structure anddistribution system is proposed to MAS agent simulation core, generated by the ASM agent,BA intelligent behavior algorithms, SRM simulation system is running, IM/OM input, outputand SSCM simulation control management sub-module consists of6ISPHT open integrationplatform/model framework, and then describe the main gathering system simulation modeling"core classes" and "constraint class" agent from the association, Multi-agent model andintegration platform relations, and key design elements ISPHT each module. On this basis, thehinterland of the port group for Cargo Agent Systems in selected booking and Cargo Portmanner, time, route choice of four decision, specific algorithms were constructed, including:subject of delivery, ports and sailing "Select Port decision model selectPort do cabinetremaining time" Cargo export"overall costs","smooth degree" with the main act"inertia"factors such constraints, by machine learning theory developed Smart Choice Harbor Qalgorithm qLearningPort(); reference collection and distribution system cost structures, theformula for estimating road transport roadTransPricing(), water transport waterTransPricing(),Railway Cargo railTransPricing() quotes module, and set transport mode decision modelselectTransMode(); analysis do learn Cargo export process Gantt chart time cabinet decisionmodel selectVanningTime();road network and the hierarchical design method, Floydalgorithm transform static line transport decision model selectStaticTransPath(), and toDijkstra’s algorithm as the core of the real-time routing update model rtTransRouteUpd().Most of the models and algorithms based on the “linear" function and rules build, effectivelycompensate for the difficult analytical modeling, logistics and limitations to considerincomplete, solving complex defects such as high, reflecting the Intelligent Agent from themicro perspective of body simulation research collection and distribution system innovativeand value.Finally, in the experimental stage, achieved by the program code ISPHT platform HighwayCargo module, on this basis, to the hinterland of the port group for the object, Dongguan City,expand the computational experiments. Experimental results show that the model can beconstructed to facilitate real-time observation and cumulative distribution of values for eachcontainer traffic to specific sections of the trunk road and is consistent with the reality of thesituation, confirmed that this article is designed to share the hinterland of the port clusterdistribution system simulation model of MAS reasonable, and simulation-based experimental study issues related to agent feasibility.更多还原