基于案例的推理在智能决策支持系统中的应用
摘要
20世纪70年代初美国人M．Ｓ．Scott Morton首先在其《管理决策系统》一文中提出了决策支持系统（Decision Support System）的概念。决策支持系统实质上是在管理信息系统和运筹学的基础上发展起来的，主要目的是面向高层次的决策，求解半结构化或非结构化的管理问题。它采用了数据库、模型库等技术，进一步扩展了决策者和计算机的人机交互功能。
决策支持系统面临着复杂的问题描述和求解，如果决策支持系统的模型库中仅有数据模型，将受到数学模型的表达能力和求解条件的限制，也不能对非公式化、非结构化的高层次决策问题提供有力的支持。因此，除了使用数学模型之外，需要采用知识模型或逻辑模型等，引入专家系统，以提高决策支持系统的问题描述能力和求解的智能水平。与此同时，由于决策者很少可能是计算机方面的专业人士，所以，决策支持系统的人机界面也需要友好、直观、方便、生动。
于是，20世纪80年代，智能决策支持系统（Intelligent Decision Support System ）就应运而生了。智能决策支持系统是决策支持系统和人工智能（Artificial Intelligent）技术相结合的产物。由于其独特的研究方法和广阔的发展前途，使之一经出现就成为决策支持系统研究的热点和主要发展方向。近20年来，智能决策支持系统的应用研究取得了巨大的进步，并且在咨询、诊断、预测、管理、设计等领域得到了广泛的应用，国内外在智能决策支持系统的理论和应用的研究方面也取得了不少成果。
但是，智能决策支持系统也还存在着一些需要解决的问题。由于大部分决策问题都是非结构化的，传统的ES采用的结构化的推理方式，无法按照描述性方式对决策过程提供智能支持。传统ES对决策支持系统各部件提供智能支持时，其学习行为大多是静态的、被动的，即按照预定的启发式策略进行学习，而不是按照实际环境需求制定动态的学习策略，缺乏主动学习机制。因此，限制了智能辅助的灵活性和适应性。此外，计算机不能正确地、及时地理解用户的提问和需求，用户不能及时获得计算机的回答和解释；用户难以对计算机进行动态干预，加入启发信息。用一般知识求解问题的方法也显示出不足。首先是知识获取瓶颈问题，除从专家那里获取知识成本高外，有时专家知道具体怎样做，但却不能有效地归纳出一般规则。其次是脆弱性，一旦处理问题所需的知识超出知识库的范围，则系统就无能为力。由于演绎过程实际上只是显示知识库中隐含的知识，而又很难保证知识库是完备的，所以脆弱性是当前基于演绎推理系统的固有缺陷。
针对上述的传统的智能决策支持系统的不足，我们引入了基于案例推理（Case-Based Reasoning）的技术方法。20世纪90年代兴起的基于案例推理（CBR）符合人类的认知心理，同时避免了规则的获取困难和不一致问题，也避免了规则提
取引起的歧义和信息丢失。因而在知识难以表达或因果关系难以把握，但已积累了丰富经验的领域，得到了广泛的应用，如法律咨询、医疗诊断、工程设计和规划等。
本文主要研究了CBR在智能决策支持系统中的应用，对案例的组织与检索、案例的改写与存储、基于CBR的IDSS模型等进行了一些探讨。首先，在对CBR技术进行探讨的基础上，研究设计了一种基于CBR的人机智能结合的智能决策支持系统模型，并对其实现的关键技术进行了讨论。其次，提出和实现了一种基于神经网络的相似案例检索算法和一种基于因果关系和规则推理的相似案例的决策方案修正算法，并将二者相结合。针对CBR中传统的案例检索模型存在的问题，我们提出了一种基于神经网络的相似案例检索算法。该算法基于三层BP网络模型，具有自组织、自适应等优点，无需定义案例属性之间的相似度和特征属性的权重。针对案例改写规则困难等问题，我们提出了一种基于因果模型与规则推理的相似案例的决策方案修正算法。它避开了案例改写规则，综合CBR、RBR和因果模型的推理机制，实现了自组织推理，提高了结论的可信度和可解释性。
本文研究的重要内容是CBR在决策支持系统中的应用，对CBR的理论和模型进行了一些探讨，为智能决策支持系统的设计和开发提供了一些新思路和实现方法。但是由于时间有限，对CBR的理论、技术和实现方法上，都还有待于进一步研究和探讨。例如：在CBR技术方面，随着CBR应用系统的运行，其案例库是不断增长的。一方面，大量求解案例的存在，为建立求解案例的案例库以及发掘案例改写规则提供了基本信息数据；另一方面，案例库可能会出现冗余、矛盾和臃肿，从而影响求解效率和精度。所以还有下列工作尚待进一步完善：（1）案例库的精简（2）建立调整案例库，即通过建立调整案例的案例库，来记录以前案例调整的经验以供将来调用。（3）利用KDD技术发掘案例改写知识。
At the beginning of the 70's of the 20th century the American M. S .Scott Morton put forward the concept of the decision support system in his 《Management Decision System》firstly. Decision Support system in fact is developing on the basis of management information system and operations research. Its main purpose is faced to high decisions. And with it we can solve half-structured and unstructured decisions in management. Decision support systems adopt the technology of database and model base so that man-machine conversation is convenient.
Decision support system is to describe and solve complicated problems. If the model base in DSS only have data model, it will be limited by the data model’s expression capability and conditions and can not provide powerful support for non-formulary or unstructured decisions. So we should adopt not only data model but also knowledge model and logic model etc. Expert system can be bring into DSS to improve problem describe capability and solving intelligence. At the same time, because the decision-maker is seldom professional in computer the interface must be friendly, intuitionistic, convenient, and lively.
Hence, in the 1980’s intelligent decision support system emerge with tide of times. IDSS is a combination of DSS and AI. For its special research method and vast developing space, it becomes the hotspot and developing direction in DSS. In the last 20 years, enormous advances have been obtained in intelligent decision support system application research. For most of the decision problems are unstructured problems, traditional reasoning method in ES can not provide intelligent support for the decision process. The learning activities are static and passive in the traditional DSS with ES. They don’t establish dynamic learning strategy according to actual conditions and lack active research mechanism. That limited the flexibility and adaptability of DSS. In addition, the computer can not accurately and duly understand asking and demands of user. As the same time, user also can not add in dynamitic inspired information. There is insufficiency in method by general knowledge. Firstly, we must face the bottleneck of knowledge obtaining. Sometimes, its cost is too high. Sometimes, though the experts know how to do and what to do, but they can not sum up common rules. Secondly, the fragility can not be avoided. Once the knowledge to handle problems is beyond scope of knowledge base, the system is incapable. For the knowledge base is not complete, so the fragility is the inherent limitation.
Aim at the shortage of the above traditional intelligent decision support system, we
bring into the technology of Case-based Reasoning. CBR that rising in 1990’s match to human perception mentality and avoid the difficulties such as obtaining rules, misunderstanding, missing of information. So in the field where knowledge is hardly to be express but a lot of experiences have been obtained, CBR have been applied widely such as the law consultation, the medical diagnosis, engineering design and planning etc.
My paper primarily researches the application of CBR in IDSS. In this paper we have discuss CBR in retrieve, reuse, revise, review, return. Firstly, we design an IDSS model based on CBR and discuss its key technique. Secondly, we put forward a similar case retrieve arithmetic based on nerve network and a similar case revise arithmetic based on Rule-Based Reasoning and causality. The former arithmetic is based on three layer BP network model and has the advantage to organize and adapt by itself. We have no need to define similarity and weigh among cases. The latter arithmetic that is based on causality and RBR is to solve the problem of rules’ revising. It has avoided rewriting rules in case, synthesizing the reasoning mechanisms of CBR, RBR and causality model. It has realized organizing reasoning by itself and improved the credibility of the conclusion and canning be explained.
Though we have researched application of CBR in IDSS especially some theories and mo
决策支持系统面临着复杂的问题描述和求解，如果决策支持系统的模型库中仅有数据模型，将受到数学模型的表达能力和求解条件的限制，也不能对非公式化、非结构化的高层次决策问题提供有力的支持。因此，除了使用数学模型之外，需要采用知识模型或逻辑模型等，引入专家系统，以提高决策支持系统的问题描述能力和求解的智能水平。与此同时，由于决策者很少可能是计算机方面的专业人士，所以，决策支持系统的人机界面也需要友好、直观、方便、生动。
于是，20世纪80年代，智能决策支持系统（Intelligent Decision Support System ）就应运而生了。智能决策支持系统是决策支持系统和人工智能（Artificial Intelligent）技术相结合的产物。由于其独特的研究方法和广阔的发展前途，使之一经出现就成为决策支持系统研究的热点和主要发展方向。近20年来，智能决策支持系统的应用研究取得了巨大的进步，并且在咨询、诊断、预测、管理、设计等领域得到了广泛的应用，国内外在智能决策支持系统的理论和应用的研究方面也取得了不少成果。
但是，智能决策支持系统也还存在着一些需要解决的问题。由于大部分决策问题都是非结构化的，传统的ES采用的结构化的推理方式，无法按照描述性方式对决策过程提供智能支持。传统ES对决策支持系统各部件提供智能支持时，其学习行为大多是静态的、被动的，即按照预定的启发式策略进行学习，而不是按照实际环境需求制定动态的学习策略，缺乏主动学习机制。因此，限制了智能辅助的灵活性和适应性。此外，计算机不能正确地、及时地理解用户的提问和需求，用户不能及时获得计算机的回答和解释；用户难以对计算机进行动态干预，加入启发信息。用一般知识求解问题的方法也显示出不足。首先是知识获取瓶颈问题，除从专家那里获取知识成本高外，有时专家知道具体怎样做，但却不能有效地归纳出一般规则。其次是脆弱性，一旦处理问题所需的知识超出知识库的范围，则系统就无能为力。由于演绎过程实际上只是显示知识库中隐含的知识，而又很难保证知识库是完备的，所以脆弱性是当前基于演绎推理系统的固有缺陷。
针对上述的传统的智能决策支持系统的不足，我们引入了基于案例推理（Case-Based Reasoning）的技术方法。20世纪90年代兴起的基于案例推理（CBR）符合人类的认知心理，同时避免了规则的获取困难和不一致问题，也避免了规则提
取引起的歧义和信息丢失。因而在知识难以表达或因果关系难以把握，但已积累了丰富经验的领域，得到了广泛的应用，如法律咨询、医疗诊断、工程设计和规划等。
本文主要研究了CBR在智能决策支持系统中的应用，对案例的组织与检索、案例的改写与存储、基于CBR的IDSS模型等进行了一些探讨。首先，在对CBR技术进行探讨的基础上，研究设计了一种基于CBR的人机智能结合的智能决策支持系统模型，并对其实现的关键技术进行了讨论。其次，提出和实现了一种基于神经网络的相似案例检索算法和一种基于因果关系和规则推理的相似案例的决策方案修正算法，并将二者相结合。针对CBR中传统的案例检索模型存在的问题，我们提出了一种基于神经网络的相似案例检索算法。该算法基于三层BP网络模型，具有自组织、自适应等优点，无需定义案例属性之间的相似度和特征属性的权重。针对案例改写规则困难等问题，我们提出了一种基于因果模型与规则推理的相似案例的决策方案修正算法。它避开了案例改写规则，综合CBR、RBR和因果模型的推理机制，实现了自组织推理，提高了结论的可信度和可解释性。
本文研究的重要内容是CBR在决策支持系统中的应用，对CBR的理论和模型进行了一些探讨，为智能决策支持系统的设计和开发提供了一些新思路和实现方法。但是由于时间有限，对CBR的理论、技术和实现方法上，都还有待于进一步研究和探讨。例如：在CBR技术方面，随着CBR应用系统的运行，其案例库是不断增长的。一方面，大量求解案例的存在，为建立求解案例的案例库以及发掘案例改写规则提供了基本信息数据；另一方面，案例库可能会出现冗余、矛盾和臃肿，从而影响求解效率和精度。所以还有下列工作尚待进一步完善：（1）案例库的精简（2）建立调整案例库，即通过建立调整案例的案例库，来记录以前案例调整的经验以供将来调用。（3）利用KDD技术发掘案例改写知识。
At the beginning of the 70's of the 20th century the American M. S .Scott Morton put forward the concept of the decision support system in his 《Management Decision System》firstly. Decision Support system in fact is developing on the basis of management information system and operations research. Its main purpose is faced to high decisions. And with it we can solve half-structured and unstructured decisions in management. Decision support systems adopt the technology of database and model base so that man-machine conversation is convenient.
Decision support system is to describe and solve complicated problems. If the model base in DSS only have data model, it will be limited by the data model’s expression capability and conditions and can not provide powerful support for non-formulary or unstructured decisions. So we should adopt not only data model but also knowledge model and logic model etc. Expert system can be bring into DSS to improve problem describe capability and solving intelligence. At the same time, because the decision-maker is seldom professional in computer the interface must be friendly, intuitionistic, convenient, and lively.
Hence, in the 1980’s intelligent decision support system emerge with tide of times. IDSS is a combination of DSS and AI. For its special research method and vast developing space, it becomes the hotspot and developing direction in DSS. In the last 20 years, enormous advances have been obtained in intelligent decision support system application research. For most of the decision problems are unstructured problems, traditional reasoning method in ES can not provide intelligent support for the decision process. The learning activities are static and passive in the traditional DSS with ES. They don’t establish dynamic learning strategy according to actual conditions and lack active research mechanism. That limited the flexibility and adaptability of DSS. In addition, the computer can not accurately and duly understand asking and demands of user. As the same time, user also can not add in dynamitic inspired information. There is insufficiency in method by general knowledge. Firstly, we must face the bottleneck of knowledge obtaining. Sometimes, its cost is too high. Sometimes, though the experts know how to do and what to do, but they can not sum up common rules. Secondly, the fragility can not be avoided. Once the knowledge to handle problems is beyond scope of knowledge base, the system is incapable. For the knowledge base is not complete, so the fragility is the inherent limitation.
Aim at the shortage of the above traditional intelligent decision support system, we
bring into the technology of Case-based Reasoning. CBR that rising in 1990’s match to human perception mentality and avoid the difficulties such as obtaining rules, misunderstanding, missing of information. So in the field where knowledge is hardly to be express but a lot of experiences have been obtained, CBR have been applied widely such as the law consultation, the medical diagnosis, engineering design and planning etc.
My paper primarily researches the application of CBR in IDSS. In this paper we have discuss CBR in retrieve, reuse, revise, review, return. Firstly, we design an IDSS model based on CBR and discuss its key technique. Secondly, we put forward a similar case retrieve arithmetic based on nerve network and a similar case revise arithmetic based on Rule-Based Reasoning and causality. The former arithmetic is based on three layer BP network model and has the advantage to organize and adapt by itself. We have no need to define similarity and weigh among cases. The latter arithmetic that is based on causality and RBR is to solve the problem of rules’ revising. It has avoided rewriting rules in case, synthesizing the reasoning mechanisms of CBR, RBR and causality model. It has realized organizing reasoning by itself and improved the credibility of the conclusion and canning be explained.
Though we have researched application of CBR in IDSS especially some theories and mo
引文
[1] 黄梯云，智能决策支持系统，电子工业出版社，2001.
[2] 高洪深，决策支持系统：理论·方法·案例，清华大学出版社，2000，第二版
[3] 陈文伟，决策支持系统及其开发，清华大学出版社，2000，第二版
[4] 蔡自兴等，人工智能及其应用，清华大学出版社，1996，第二版
[5] Joseph Giarratano，专家系统原理与编程，机械工业出版社，2000
[6] Ralph M.Srair，信息系统原理，机械工业出版社，2000
[7] 胡运权，运筹学教程，清华大学出版社，2003，第二版
[8] 刘怀亮等，智能决策支持系统的模型技术及研究趋势，工业控制计算机，2003，16(2)
[9] 李楠等，决策支持系统中粗糙集理论的应用研究，计算机工程，2003，29(1)
[10] 尹春华等，决策支持系统研究现状及发展趋势，决策借鉴，2002，15(2)
[11] 胡亚辉等，CBR技术中事例表达的研究与应用，机械工程师，2003，2
[12] 黄浩等，基于CBR系统事例检索算法的研究，计算机应用，2003，23(6)
[13] 许志瑞等，一种基于神经网络的事例相似度评估方法，华中理工大学学报，2000，28(10)
[14] 江勤等，基于CBR专家系统案例知识的检索、匹配及其扩展，山东科技大学学报，2002，21(2)
[15] 段军等，基于Rough Set和神经网络的CBR快捷检索方法，计算机工程与应用，2003，3
[16] Yuh Foong David Law etc，An integrated case-based reasoning approach for intelligent help desk fault management，Expert System with Applicat-
ion，1997，13(4)
[17] Jarmulak J etc，Case-based reasoning in an ultrasonic rail-inspection system 1997,Second International Conference on Case-Based Reasoning
[18] Bach C，Case-based reasoning in diagnostic expert system，AI Communicati-
on，1996，9(2)
[19] MacCarthy etc，Case-based reasoning in scheduling，Mech.Eng.Publication
1996
[2] 高洪深，决策支持系统：理论·方法·案例，清华大学出版社，2000，第二版
[3] 陈文伟，决策支持系统及其开发，清华大学出版社，2000，第二版
[4] 蔡自兴等，人工智能及其应用，清华大学出版社，1996，第二版
[5] Joseph Giarratano，专家系统原理与编程，机械工业出版社，2000
[6] Ralph M.Srair，信息系统原理，机械工业出版社，2000
[7] 胡运权，运筹学教程，清华大学出版社，2003，第二版
[8] 刘怀亮等，智能决策支持系统的模型技术及研究趋势，工业控制计算机，2003，16(2)
[9] 李楠等，决策支持系统中粗糙集理论的应用研究，计算机工程，2003，29(1)
[10] 尹春华等，决策支持系统研究现状及发展趋势，决策借鉴，2002，15(2)
[11] 胡亚辉等，CBR技术中事例表达的研究与应用，机械工程师，2003，2
[12] 黄浩等，基于CBR系统事例检索算法的研究，计算机应用，2003，23(6)
[13] 许志瑞等，一种基于神经网络的事例相似度评估方法，华中理工大学学报，2000，28(10)
[14] 江勤等，基于CBR专家系统案例知识的检索、匹配及其扩展，山东科技大学学报，2002，21(2)
[15] 段军等，基于Rough Set和神经网络的CBR快捷检索方法，计算机工程与应用，2003，3
[16] Yuh Foong David Law etc，An integrated case-based reasoning approach for intelligent help desk fault management，Expert System with Applicat-
ion，1997，13(4)
[17] Jarmulak J etc，Case-based reasoning in an ultrasonic rail-inspection system 1997,Second International Conference on Case-Based Reasoning
[18] Bach C，Case-based reasoning in diagnostic expert system，AI Communicati-
on，1996，9(2)
[19] MacCarthy etc，Case-based reasoning in scheduling，Mech.Eng.Publication
1996