《基于案例推理的应急管理案例库构建方法研究》由会员分享,可在线阅读,更多相关《基于案例推理的应急管理案例库构建方法研究(74页珍藏版)》请在金锄头文库上搜索。
1、上海交通大学硕士学位论文基于案例推理的应急管理案例库构建方法研究姓名:王凯申请学位级别:硕士专业:企业管理指导教师:季建华20090101-I- 基于案例推理的应急管理案例库构建方法研究 摘 要 在企业缩减供应链成本、提高供应链效率的压力下,供应链纵向上的企业越来越多, 且分布的地域更加广泛, 而供应链横向上的同质可替代供应商、 制造商、分销商越来越少,促使供应链向窄长化发展,从而大大增加了供应链的脆弱性和企业运作中的不确定性。而随着人类经济活动范围的持续扩大和深度的不断推进,环境破坏、气候变化等引起的自然灾害和各种恐怖袭击、罢工、地区冲突等社会公共事件不断出现,这些事件的出现往往具有突发性、
2、异常性、急迫性、信息高度缺失性等特点。 突发事件的频发性与供应链的脆弱性之间的矛盾已经成为日渐突出的问题, 对突发事件的分类研究与应对也是政府和企业都日益重视的管理内容。 我国公共管理领域对突发事件一般分为自然灾害、 事故灾难、 公共卫生事件、社会安全事件等四大类,并分别有国家级和省级的不同预案,每类事件按照严重程度的差异由不同级别的行政部门负责处理。 而企业管理领域尚未有严格的突发事件管理方法,较为常见的是对企业运作和经营风险的监控、预警和处理预案,这些一般风险也具备突发性和急迫性的特点,但与突发事件相比,还缺少异常性和信息高度缺失性的特点,因此这些管理和分类方法只能作为参考。公共管理领域与
3、企业管理领域的突发事件分类方法既有重叠又有不相容之处, 二者的融合有利于扩大分类方法的通用性。由于领域知识不完备,推理规则不明确,应急管理领域迫切需要实际案例为基础的管理应用和规则提取机制, 基于案例推理的方法则可以同时满足这两个方面的需求。 基于此,根据突发事件发生的范围,从企业视角出发提出了突发事件分级分类方法,将突发事件分为突发自然灾害、突发公共事件、供应链突发事件及企业突发事件,全面覆盖企业运作中可能面临的各类事件,同时兼顾公共管理领域的需求,将现有政府分类方法中的四类突发事件包含其中。 使用基于案例推理的方法建立应急管理案例库, 最基本和最重要的步骤有案例的表示、案例属性的约简、案例
4、的索引与检索以及案例的修改与学习等。框架表示法以其结构性、 继承性、 自然性的优点, 尤其适用于对复杂案例知识的表示,能够将同时包含自然语言文本描述和各种定量数据信息的案例知识进行结构清晰、层次分明的结构化表示,并用子框架嵌套的形式表示案例之间的继承关系。当案例数目逐渐增多时,为了提高系统效率,需要对多余的属性进行约简,以提上海交通大学硕士学位论文 基于案例推理的应急管理案例库构建方法研究 -II- 高案例表示及检索的效率。 对案例表示中的冗余属性进行基于信息熵的约简有较好的效率和可操作性,该算法通过熵的大小衡量不同属性对案例的分类能力,从而判断不同属性对推理的影响大小,再结合供应链管理和应急
5、管理的已有知识,即可考虑删除权重较小的属性。 为了提高案例检索的效率和准确度,案例检索方法结合了分类索引、关键字检索、KNN 算法等,能够分别处理具有明显分类能力的案例属性、文本描述属性以及定量属性,使每类属性值在检索中达到最高利用率。其中,KNN 算法中的属性权重初始值由专家打分结合 AHP 方法获得,在案例库扩大之后,采用神经网络算法对属性权重进行基于案例本身信息的调整, 以更好的挖掘案例本身包含的隐性知识。对案例的修改和学习则主要采用重实例化和诱导重放的方法,以更好的利用已有体系结构学习新的知识,提高案例库的整体性能。 对案例的存储和系统的初步实现采用 MS Access 单机数据库,
6、重点实现案例表示与案例检索的关键模块。 案例知识的表示及存储通过 6 个基本数据表和其他辅助功能表完成,案例的检索则主要通过 SQL 查询和 KNN 计算的结合来实现。 关键词:关键词: 突发事件,基于案例推理,框架表示法,属性约简,KNN 算法 -III- RESEARCH ON DISRUPTION MANAGEMENT CASE DATABASE CONSTRUCTION WITH CASE-BASED REASONING ABSTRACT With increasing pressure to reduce cost and improve supply chain efficienc
7、y, there are more enterprises in supply chain spreading in larger regions. Simultaneously, there are less substitutive suppliers, manufacturers and distributers on horizontal supply chain nodes. As a result, the supply chains are becoming thinner and longer, which highly contributes to the vulnerabi
8、lity and uncertainty of supply chains. On the other hand, human beings economic activities have further extended and deepened in recent years. Environmental problems and pollutions are becoming fiercer and terrorism, workers strike and regional conflict are becoming more frequent. These disruptions
9、are often sudden, abnormal, urgent and seriously lacking necessary information to deal with. The confliction between frequent disruptions and supply chain vulnerability is becoming fiercer nowadays, making both government and enterprises emphasize more on disruption classifying and management. Disru
10、ptions in the public management scope are divided into four categories natural disasters, accidents, public health incidents and social security incidents, for the convenience of supervision and management from government departments of different levels and various functions. In contrast, disruption
11、 management in business administration area is still relatively less emphasized. Similar practices are risk management, including the monitoring, warning and handling of operational risks. Risk management also needs to deal with sudden and urgent emergencies, but these risks are usually not abnormal
12、 and seriously lacking information. The traditional disruption categories in government and enterprise management fields are overlapped and controversial, which arouses the need of a more general disruption classifying methodology. Considering the significant lack of knowledge and rules in this fiel
13、d, case-based reasoning method is especially beneficial for disruption management. Accordingly, based on the scope of disruptions breaking out and enterprises view point, disruptions can be classified into four categories gusty natural disasters, 上海交通大学硕士学位论文 基于案例推理的应急管理案例库构建方法研究 -IV- gusty public i
14、ncidents, supply chain disruptions and enterprise disruptions. Thus the various disruptions in enterprises business and government administration can be included. The fundamental and key steps in applying CBR are case indicating, case attributes reduction, case indexing and searching, case revising
15、and learning. Frame method is applied to organize cases in clear structures, due to its structural, inheriting and natural style features. Information and knowledge in natural language and quantitative data can both be indicated with frame method, applying its inheriting feature between father-frame
16、 and child-frame. Case attributes need to be reduced to enhance the system efficiency when case amount is sharply increasing. Reluctant attributes would be contracted with a method based on information entropy. Entropy is applied to indicate the case classifying capability of different attributes, showing the relative importance of current attributes. Consequently, attributes of least importance could be removed after a ca