【作者】 詹蓉；

【导师】 陈荣秋；

【作者基本信息】 华中科技大学 ， 管理科学与工程， 2008， 博士

【摘要】 近年来,一种新的被称为“即时顾客化定制”(下文简称IC)的制造模式引起了学术界和实际工作者的重视。IC是指一旦顾客提出个性化需求,制造商能立即交付正确的产品。IC能同时实现定制、零顾客订货提前期和低成本的目标。为了实现IC,一个重要的运作策略是按个性化需求预测生产。IC使得个性化需求预测更加重要,尤其是短期预测。但是,现有文献中至今没有发现通用的个性化需求定量预测方法的研究。因此,本文研究的内容属于一个新的领域,其特色在于开创性。本文从定量的角度深入地研究了个性化需求预测问题,得到了如下创新成果。首先,提出了个性化需求的概念、特点和定性分类方法。现有文献缺乏对个性化需求概念的明确定义,本文提出广义的个性化需求和狭义的个性化需求的概念,并提出产品定制属性的概念。接着全面地研究了个性化需求的主要特点。按不同标准对个性化需求进行定性分类,分类方法包括按消费者购买的基本动机分类,按个性化需求与定制产品设计、制造的关系分类,按消费者心理学分类,按市场营销学分类。第二,提出个性化需求定量分类和聚类的方法。现有文献中没有发现关于个性化需求定量分类和聚类的研究以及相应的具体方法。本文根据不同的标准对个性化需求进行了定量分类,这些标准包括个性化需求的相关性、不同客户需求数量的均匀程度、个性化需求的自相关性以及个性化需求交付频率的高低。对每种分类标准提出了定量分类的详细步骤和方法。然后提出基于主成分分析的个性化需求聚类的步骤。第三,提出一个基于案例推理的个性化需求预测方法。现有文献中没有发现通用的个性化需求预测方法。个性化需求预测的难点和重点在于不同类型属性的值必须同时预测出来,预测非数值属性尤其困难。对此,本文提出一个面向属性的预测思路,以便个性化需求预测可转变为数值预测。还提出数据预处理的详细过程,使个性化需求预测转变为数值型时间序列预测问题。接着,本文提出一个基于案例推理的系统来进行通用的个性化需求预测。它分为三个阶段:案例的表达、相似性搜索、案例的预测和调整。尤其是,本文提出广义的时间序列距离计算公式,将包含时间序列片段的斜率和长度的数据对来描述时间序列的变化特征。该公式形式简单,又具有高度的柔性,应用时可繁可简,可以将目前很多相关的方法统一在一个框架中,从而丰富、拓展了时间序列相似性度量的方法。在某种程度上,王达等提出的时间序列的模式距离可以认为是本文所提出公式的特例。本文提出的方法具有普遍意义,还可应用在其它很多领域。然后用一家酒店的真实数据评估了预测绩效,实践结果证明本文提出的方法能够提供高精度的预测。最后,提出了个性化需求预测支持系统中的其他预测方法。首先分析了个性化需求预测的特点,并拓展了个性化需求预测的概念,分析了个性化需求预测支持系统的要求。接着研究马尔可夫链在个性化需求预测中的实际应用。然后针对个性化需求预测的特点,提出相应的组合预测方法。对相关子集的预测提出一些基本的思路和方法。更多还原

【Abstract】 In recent years, much attention has been focused by practitioners as well as academics on a new manufacturing paradigm called ’instant customerisation’(IC). IC means that once customer’s individual demand is put forward, a manufacturer can deliver the right product immediately. IC can realize the synergies among customisation, zero customer lead time, and low cost. One of critical operation tactics is finalise-to-individual demand forecasting to realize IC. IC attaches more importance to individual demand forecasting, especially short-term forecasts. However, there is scarcely research on universal quantificational individual demand forecasting method until now. Therefore the content in the dissertation belongs to a new field, and its characterstic is originality. An in-depth study for forecasting individual demand quantificationally results in innovative production as follows.Firstly, the concept, characteristic and qualitative classified methods are proposed. There is scarcely research about the concept of individual demand. We propose the concept of individual demand with broad sense and narrow sense respectively. In addition, the concept of product customerisation attributes is proposed. Then the main character of individual demand is studied and individual demand is classified according to various criteria. These criteria include basic purchasing motivation of consumer, the relation between individual demand and customisation product design or manufacturing, consumer psychology and marketing.Secondly, quantificational classified and clustering methods of individual demand are proposed. There is hardly any such research. According to different criterion, individual demands are classified quantificationally. These criteria include demands’ correlation, well-proportioned extent of different customers’ demand quantity, demands’ autocorrelation and delivering frequency. Detailed steps and methods of quantificational classification are described for every criterion. Then, the method of individual demand clustering based on principal component analysis is proposed.Thirdly, we propose a case-based reasoning method for individual demand fore- casting. There is scarcely research on universal individual demand forecasting method. An important feature and difficulty of individual demand forecasting is that the value of different types of attributes must be forecasted at the same time. It is especially difficult to forecast nonnumeric attributes. Then an oriented-attribute forecasting approach is proposed in order that individual demand forecasting can be transformed into numeric forecasting. We describe how the data is preprocessed in detail, therefore individual demand forecasting can be translated into numeric time series forecasting. Then we propose a case-based reasoning (CBR) system for universal individual demand forecasting. The CBR system is divided into three stages: case representation, similarity search, and case forecasting and adjustment. In particular, we propose a distance-based formula that broadens our view by regarding the slope and length of a segment as changing characteristics of a time series. Basing on the compactness and flexibility of the formula form, its application can be either complicated or simple in various environments. Since many relevant methods can be unified into a framework by the use of our formula, it enriches and expands the measure method of time series. To some extent, pattern distance of time series proposed by Wang Da et al. is a special situation of our formula. As a universal method, it is able to be applied in many other fields. The performance of our system is evaluated using real data from a restaurant. The empirical results show that our method is able to produce a forecast with a high degree of accuracy.At last, other individual demand forecasting method in the forecasting support system is studied. The character of individual demand forecasting is analyzed, and the concept of individual demand forecasting is expanded. The request of individual demand forecasting support system is proposed. Then the application of markov chains for individual demand forecasting is discussed. The corresponding combination forecasts methods are proposed according to the character of individual demand forecasting. Some basic approach and methods about correlation sub-aggregate are put forward.更多还原