节点文献
置信规则推理方法及其在库存与生产运作管理中的应用
Belief-Rule Inference Method and its Applications in Inventory and Production Operations Management
【作者】 李彬;
【作者基本信息】 华中科技大学 , 系统工程, 2012, 博士
【摘要】 随着信息与制造技术的加速革新和产品生命周期的缩短,企业及供应链管理中的库存与生产运作管理环节面临着越来越多的不确定性,其中最大的来源是需求的不确定性。系统地分析和解决不确定性需求下的库存与生产运作管理问题成为当今研究学者和实践者研究的热门课题。解析式和仿真式方法在求解复杂性、移植性、领域知识和不确定信息的表达能力等方面存在局限。启发式方法可以有效利用专家领域知识并适用于各种复杂和变化的资源约束环境下,特别是人工智能启发式方法还可以良好地表达和处理各种不确定信息。面对不断动态突变的库存与生产运作环境,伴随着频繁的重新规划和关联的资源约束,研究学者和实践者越来越趋于应用人工智能启发式方法。置信规则推理方法是在证据推理理论和产生式规则专家系统的基础上提出来的解决不确定性问题的人工智能启发式方法。本文对置信规则推理理论进行深入研究,并应用于不确定需求下的库存与生产运作管理问题,主要研究工作如下:论述了置信规则推理理论的基础知识,总结出置信规则推理方法的应用模式和应用机制,提出在区间不确定性信息输入下的置信规则前向推理算法。将置信规则推理方法的应用模式分为系统逼近器和系统控制器。应用机制提供了置信规则推理方法的关键环节,为置信规则推理方法的改进和应用提供了指导和支撑作用。针对不确定需求下的库存控制问题,分别考虑缺货补偿和缺货丢失两种情况,提出基于置信规则推理的库存控制方法。证明了正态随机预测误差下的最优基本库存策略作为定量专家知识初始化置信规则库,为定性专家知识缺失或不可信时提供良好的依据。给出一个数值案例和一个4S店实例进行仿真验证,并与现有的适应性、启发式、近似最优、确定等价控制和鲁棒优化等方法进行比较分析。针对置信规则库的结构识别和参数识别,提出了同步识别和异步识别方法。在异步识别中提出置信K均值聚类算法对置信规则库的结构进行识别。对于不确定需求信息下的企业集约生产规划问题,提出基于置信规则推理的层次化集约生产规划方法,并构建了连续和切换生产模式。给出某机动车和匹兹堡油漆厂两个生产规划案例,将同步识别和异步识别方法进行了比较分析,并将置信规则推理方法与非线性区间数规划、线性决策规则和生产切换启发式方法进行比较,同时提供了针对不同成本结构下置信规则推理方法的敏感性分析。针对区间信息输入下的置信规则前向推理算法、异步识别中的结构识别和参数识别算法,分别提出了相应的遗传共轭梯度求解方法,并应用于不确定需求情况下的集中式供应链管理问题。对于由单个生产商和单个分销商组成的两级串型供应链系统,提出生产商和分销商的层次化置信规则推理框架,构建了生产商和分销商的置信规则推理订货策略。在某机动车制造商和分销商组成的两级串型供应链实例中分析了遗传共轭梯度求解方法相对于已有求解方法的优越性,并将置信规则推理方法与鲁棒优化方法进行了比较。
【Abstract】 With the rapid development of information and manufacturing technologies andgradually shortened product life cycle, inventory and production operations management inenterprise and supply chain management are facing more and more uncertainties, in whichthe biggest source is demand uncertianty. It has become a hot spot nowadays whichattracted much attention from researchers and practitioners to systematically analyze andsolve inventory and production operations management problems under uncertainties. Theanalytical method and simulation method have limitations with respects to computationalcomplexity, portability, and the expression ability of domain knowledge and uncertaininformation. Heuristic methods can effectively use expert domain knowledge and aresuitable for complex, changing and resource constrained environment, especially theartificial intelligence based heuristic methods have favorable capability to express andtransact uncertainty. Facing dynamically changing inventory and production operationalenvironments with frequent replanning and interconnected resource constraints, researcherand practitioner more and more tend to use AI based heuristic methods.As an AI based heuristic method, belief-rule-based inference (BRBI) method isproposed based on evidential reasoning theory and production rule based expert system fordealing with uncertain problems. This thesis studies deeply the BRBI method, and applies itto inventory and production operations management problems under uncertainty. The mainresearch works are as follows:This thesis introduces the foundation of BRBI theory, outlines the application modesand application mechanisms of BRBI method, and proposes a BRBI approach with intervaluncertain inputs. The application modes of the BRBI model are classified into systemapproximator and system controller. The application methanisms provide the key elimentsof the BRBI model, which has instruction and support functions to improvement andapplication of the BRBI method.For inventory control problem under nonstationary uncertain demand, abelief-rule-based inventory control (BRB-IC) method is proposed considering bothbackorder case and lost sales case. An optimal base stock policy under normal forecast erroris proved as a quantitative expert knowledge to initialize the belief-rule-base. A numericalexample and an auto4S store case study are provided to examine the BRB-IC method bycomparison with existing adaptive method, heuristic method, myopic method,certainty-equivalent-control, and robust optimization.For structure and parameter identifications of the belief-rule-base, a simultaneous identification approach and an asynchronous identification approach are developed andcompared. In the asynchronous approach, a belief K-means clustering algorithm is putforward for structure identification. For aggregate production planning (APP) underuncertain demand, a hierarchical BRBI method for APP is proposed including bothcontinuous and switching modes. The simultaneous identification approach andasynchronous identification approach are applied to identify the structure and parameter ofBRB. An automotive case study and the Pittsburgh paint factory example are provided. TheBRBI method is compared with nonlinear interval number programming, linear decisionrule and production switching heuristic. The sensitivity of BRBI method is analyzed underdifferent cost structures.For belief-rule-inference model under interval inputs, structure identification andparameter identification models in the asynchronous approach, corresponding genetic-conjugate gradient algorithms are proposed and applied into centralized supply chainmanagement problem under uncertain demand. The producer and distributor’s hierarchicalBRBI framework is constructed and their inference models and order policies are provided.In an automotive case study, the superiority of the genetic-conjugate gradient algorithmsover existing algorithms are provided, and the BRBI method is compared with the robustoptimization method.