基于混合型专家系统的企业信用评估研究
|
摘要
信用评价是对企业能否如约还本付息的能力和可信任程度的评估。借助评价模型对企业的信用作出准确评价和判断，极具社会和经济价值。但是企业信用评价又是一个十分复杂的过程，要综合考虑企业各方面的情况，同时包括定量和定性两方面的分析。传统的信用评价模型有财务比率分析、多元判别分析等，但这些模型都不能有效、全面、完整地解决信用评价问题。
随着人工智能技术的发展，已有学者提出将神经网络技术应用于信用评价领域，并取得了令人鼓舞的结果。但是在实际的应用中，神经网络技术并不能完整描述整个信用评价过程。本文提出了基于混合型专家系统的信用评价技术，利用神经网络处理财务状况方面的定量分析，利用专家系统来完成定性指标处理、数据预处理、综合企业情况判定企业信用等级等功能。这种技术符合信用评价的特点，更适合于建立信用评价模型。
本文主要工作是建造专家系统部分。以某银行提供的企业样本数据为基础，在完成对样本数据处理、抽取财务比率等基本工作后，根据信用评价领域知识的特点构建了系统的知识库，设计了基于优先级的正向推理机制与基于事实的自动解释机制，然后采用面向对象技术对系统进行分析，建立专家系统的对象模型和功能模型，在此基础上，采用access数据库设计系统的知识库，采用C++ Builder开发软件实现了基于混合型专家系统的可视化信用评价系统原型。实验结果表明，混合型专家系统既具有专家系统灵活性、解释性的特点，又具有神经网络鲁棒性、自学习能力的特点，因此非常适合用于企业信用评价，具有广泛的应用前景。
本文具体内容安排如下：
第一章介绍了信用、信用风险、信用评价的概念，回顾了信用评价的历史、发展和现状，并综合各种信用评价模型，指出这些模型各自的优缺点：
第二章简单描述了人工智能技术，着重介绍有关专家系统与神经网络的基础知识，通过总结它们的优缺点，指出结合专家系统与神经网络构造混合型专家系统的必要性；本章还介绍了神经网络子模块的概念，提出了混合型专家系统的一般框架与设计步骤：
第三章对样本数据进行处理，包括异常数据的剔除、因子分析等，提出了信用评价混合型专家系统的具体框架结构，介绍了系统知识库的主要部分、基于优先级的正向推理机制的流程、以及基于事实的自动解释机制的具体实现方法；
第四章介绍了面向对象技术，进而采用面向对象对信用评价系统进行分析，建立了对象模型和功能模型，并在此基础上，采用C++语言以规则类为例说明系统中具体类的实现，用伪代码的形式描述了推理的算法；
第五章描述了整个系统的结构，对系统主要功能模块和界面进行了介绍，并总结系统的特点；
第六章总结了全文，指出本文所构造系统存在的不足以及对将来的展望。
Credit assessment is the evaluation of the likelihood for enterprise to repay their loans and interests. It is very important and valuable to evaluate enterprise properly with the assistance of credit assessment models. However,including both quantitative and qualitative analysis,credit assessment is a rather complicated procedure in which many aspects should be considered. Thus,traditional credit assessment models,such as financial ratio analysis,multiple discriminate analysis and so on,cannot solve this problem effectively,completely and perfectly.
With the development of AI,there are scholars who apply neural network to credit assessment and have already got some promising results. But in practice,neural network can still not illustrate the whole credit assessment process perfectly. In this paper,a new method based on mixed-expert system is introduced. In the prototype system constructed in this paper,we use rule-based expert system to deal with qualitative factors and neural network to trait quantitative data. Thus,a mixed-expert system is more suitable to the enterprise credit assessment.
This work is contributed to the design and the construction of the traditional expert system part of the mixed system. Firstly,we deal with sample data provided by a bank in Fujian and abstract financial ratios by statistical methods;secondly,we construct knowledge base according to area-knowledge of credit assessment,and design a priority-based forward-inference engine and a fact-based automatic explaining mechanism;thirdly,we adopt object-oriented technology to analysis the system and construct object model and functional model of expert system;finally,by using ACCESS database to construct system knowledge base,and C++ Builder software to program,we develop a credit assessment prototype system based on mix-expert system. The experimental results show that the system constructed by this method has the virtues of robustness,flexibility,explicable ability and self-study capability,since the new method has got both the merits of expert system and of neural network. The content of this paper is arranged as foll
owing:
Chapter 1 introduces the concept of credit,credit risk and credit assessment,as well as the history and development of credit assessment;
Chapter 2 introduces the history of AI technology,and the background of expert system and neural network. Characters and disadvantages of expert system and neural network are presented respectively and the necessity of combining expert system and neural network is lightened;
Chapter 3 shows the process of dealing with sample data,including the treatment of exceptional data and factor analysis,and puts forward the concrete framework of the mixed-expert credit assessment system;
Chapter 4 introduces concept of object-oriented technology,and constructs object model and functional model after analyzing the whole system. It also illustrates the implementation of concrete classes by an example of rule class and the inference algorithm in the form of pseudocode;
Chapter 5 introduces the structure of the whole system,the major functional models and their interfaces,and the characteristic of the system is also generalized;
Chapter 6 summarizes the whole work,and points out the remaining deficiencies as well as the prospective of this method.
随着人工智能技术的发展，已有学者提出将神经网络技术应用于信用评价领域，并取得了令人鼓舞的结果。但是在实际的应用中，神经网络技术并不能完整描述整个信用评价过程。本文提出了基于混合型专家系统的信用评价技术，利用神经网络处理财务状况方面的定量分析，利用专家系统来完成定性指标处理、数据预处理、综合企业情况判定企业信用等级等功能。这种技术符合信用评价的特点，更适合于建立信用评价模型。
本文主要工作是建造专家系统部分。以某银行提供的企业样本数据为基础，在完成对样本数据处理、抽取财务比率等基本工作后，根据信用评价领域知识的特点构建了系统的知识库，设计了基于优先级的正向推理机制与基于事实的自动解释机制，然后采用面向对象技术对系统进行分析，建立专家系统的对象模型和功能模型，在此基础上，采用access数据库设计系统的知识库，采用C++ Builder开发软件实现了基于混合型专家系统的可视化信用评价系统原型。实验结果表明，混合型专家系统既具有专家系统灵活性、解释性的特点，又具有神经网络鲁棒性、自学习能力的特点，因此非常适合用于企业信用评价，具有广泛的应用前景。
本文具体内容安排如下：
第一章介绍了信用、信用风险、信用评价的概念，回顾了信用评价的历史、发展和现状，并综合各种信用评价模型，指出这些模型各自的优缺点：
第二章简单描述了人工智能技术，着重介绍有关专家系统与神经网络的基础知识，通过总结它们的优缺点，指出结合专家系统与神经网络构造混合型专家系统的必要性；本章还介绍了神经网络子模块的概念，提出了混合型专家系统的一般框架与设计步骤：
第三章对样本数据进行处理，包括异常数据的剔除、因子分析等，提出了信用评价混合型专家系统的具体框架结构，介绍了系统知识库的主要部分、基于优先级的正向推理机制的流程、以及基于事实的自动解释机制的具体实现方法；
第四章介绍了面向对象技术，进而采用面向对象对信用评价系统进行分析，建立了对象模型和功能模型，并在此基础上，采用C++语言以规则类为例说明系统中具体类的实现，用伪代码的形式描述了推理的算法；
第五章描述了整个系统的结构，对系统主要功能模块和界面进行了介绍，并总结系统的特点；
第六章总结了全文，指出本文所构造系统存在的不足以及对将来的展望。
Credit assessment is the evaluation of the likelihood for enterprise to repay their loans and interests. It is very important and valuable to evaluate enterprise properly with the assistance of credit assessment models. However,including both quantitative and qualitative analysis,credit assessment is a rather complicated procedure in which many aspects should be considered. Thus,traditional credit assessment models,such as financial ratio analysis,multiple discriminate analysis and so on,cannot solve this problem effectively,completely and perfectly.
With the development of AI,there are scholars who apply neural network to credit assessment and have already got some promising results. But in practice,neural network can still not illustrate the whole credit assessment process perfectly. In this paper,a new method based on mixed-expert system is introduced. In the prototype system constructed in this paper,we use rule-based expert system to deal with qualitative factors and neural network to trait quantitative data. Thus,a mixed-expert system is more suitable to the enterprise credit assessment.
This work is contributed to the design and the construction of the traditional expert system part of the mixed system. Firstly,we deal with sample data provided by a bank in Fujian and abstract financial ratios by statistical methods;secondly,we construct knowledge base according to area-knowledge of credit assessment,and design a priority-based forward-inference engine and a fact-based automatic explaining mechanism;thirdly,we adopt object-oriented technology to analysis the system and construct object model and functional model of expert system;finally,by using ACCESS database to construct system knowledge base,and C++ Builder software to program,we develop a credit assessment prototype system based on mix-expert system. The experimental results show that the system constructed by this method has the virtues of robustness,flexibility,explicable ability and self-study capability,since the new method has got both the merits of expert system and of neural network. The content of this paper is arranged as foll
owing:
Chapter 1 introduces the concept of credit,credit risk and credit assessment,as well as the history and development of credit assessment;
Chapter 2 introduces the history of AI technology,and the background of expert system and neural network. Characters and disadvantages of expert system and neural network are presented respectively and the necessity of combining expert system and neural network is lightened;
Chapter 3 shows the process of dealing with sample data,including the treatment of exceptional data and factor analysis,and puts forward the concrete framework of the mixed-expert credit assessment system;
Chapter 4 introduces concept of object-oriented technology,and constructs object model and functional model after analyzing the whole system. It also illustrates the implementation of concrete classes by an example of rule class and the inference algorithm in the form of pseudocode;
Chapter 5 introduces the structure of the whole system,the major functional models and their interfaces,and the characteristic of the system is also generalized;
Chapter 6 summarizes the whole work,and points out the remaining deficiencies as well as the prospective of this method.
引文
[1] 马歇尔，叶元龙，郭家麟．货币、信用与商业．商务印书馆，1986
[2] 喻瑞祥．货币信用与银行．中国财政经济出版社，1980
[3] 吴晓灵，李德。金融业的风险管理与信用评估．中国金融出版社，1996
[4] 吴晶妹。中国企业与证券的资信评估．中信出版社，1994．6
[5] 吉晓辉，黄纪宪．国有商业银行的贷款管理．同济大学出版社，1993．12，108
[6] 财政部统计评价司．企业效绩评价问答．经济科学出版社，1999
[7] Lyn C.Thomas. A survey of credit and behavioural scoring: forecasting financial risk of lenging to consumers. International Journal of Forecasting(2000) 149-172
[8] William F. Treacy, Mark Carey, Federal Resserve Board. Credit risk rating systems at large US banks, Journal of Banking & Finance 24(2000), 167-201
[9] Michel Crouhy, Dan Galai,Robert Mark. Prototype risk rating system. Journal of Banking & Finance 25(2001)47-95
[10] A fuzzy credit-rating approach for commercial loans: a Taiwan case,Liang-Hsuan Chen,Tai-Wei Chiou,Department of Industrial Management Science,National Cheng Kung University, Tainan,Taiwan,ROC, the International Journal of Management Science(1999),407-419
[11] 叶武．基于人工神经网络的银行客户信用评价系统的研究．厦门大学自动化系硕士学位论文，2000
[12] Masanori Amano. Credit rationing of a Bayesian bank with simple screening technologies. Japan and the World Economy11(1999)545-556
[13] Erkki K.Laitinen, Prediting a corporate credit analyst's risk estimate by logistic and linear models.international Review of financial Analysis(1999) 97-121
[14] 张维，李玉霜，王春峰．递归分类树在信用风险分析中的应用．系统工程理论与实践，2000年第三期
[15] 王春峰，康莉．基于遗传规划方法的商业银行信用风险评估模型．系统工程理论与实践，2001年第二期
[16] 王春峰，万海晖，张维．基于神经网络技术的商业银行信用风险评估．系统工程理论与实践，1999．9
[17] David West. Neural network credit scoring models. Computers & Operations Research(2000) 1131-1152
[18] Romaniuk SG, Hall LO.Decisionmaking on credit-worthiness,using a fuzzy connectionist model.Fuzzy Sets and Systems 1992;48:15-22
[19] Michael B.Gordy. A comparative anatomy of credit risk models. Journal of Banking & Finance 24(2000) 119-149
[20] S.Kanungo, S.Sharma, P.K.Jain. Evaluation of a decision support system for credit management decisions. Decision Support Systems30 (2001) 419-436
[21] 王万森．人工智能原理及其应用．电子工业出版社，2000．9
[22] 蔡自兴，徐光佑．人工智能及其应用．清华大学出版社，1996
[23] 田盛丰，黄厚宽．人工智能与知识工程．中国铁道出版社，1998
[24] 尹朝庆，尹浩．人工智能与专家系统．中国水利水电出版社，2001
[25] 杨炳儒．知识工程与知识发现．冶金工业出版社，2000
[26] 曹文君．知识库系统原理及其应用．复旦大学出版社
[27] 施鸿宝，王秋荷．专家系统．西安交通大学出版社
[28] 程慧霞等．用C++建造专家系统．电子工业出版社，1996
[29] 石纯一等．基于解释的机器学习方法．清华大学出版社，广西科学技术出版社，1997
[30] 汪培庄，李洪兴．知识表示的数学理论．天津科学技术出版社，1994
[31] 刘叙华．基于归结方法的自动推理．科学出版社，1994
[32] 王耀南．智能控制系统 模糊逻辑专家系统神经网络控制．湖南大学出版社
[33] 郭志刚．社会统计分析方法——-SPSS软件应用．中国人民大学出版社，1999．12
[34] 卢纹岱，朱一力，沙捷，朱红兵．SPSS for Windows从入门到精通．电子工业出版社，1999
[35] W. Greene,Yasuhiro Sakai. Comments on "Sample selection in credit scoring models". Japan and the World Economy(1998) 317-320
[36] Differences of opinion and selection bias in the credit rating industry, Richard Cantor, Frank Packer ,Federal Reserve Bank of New York,33 Liberty Street, New York,NY 10045,USA, Journal of Banking & finance 21(1997) 1395-1417
[37] 王林树，鲍兰平，赵瑞清．概率论与数理统计．科学出版社，1999．8
[38] 陈雄华．基于神经网络和专家系统结合的企业信用评估研究．厦门大学自动化系硕士学位论文，2002．6
[39] 高严．如何阅读三张表．经济科学出版社，2000．10
[40] Chih-Chou Chiu, Ling-Jing Kao. Combining a neural network with a Rule-Based Expert System.Approach for short-term Power Load Forecasting in Taiwan.Expert Systems with Applications.vol 13 No.4(299-305) 1997
[41] Zhang Liuyuan Lin Chengde. A New Method for Credit Assessment in Commercial Bank. Proceedings of the International Conference on Modeling and simulation in distributed applications, 2001.9,804-808
[42] 钟佩思 高国安．基于神经网络模块的混合型方案设计知识库系统．机械设计 No．5，1995．5
[43] M.S.Kandil, S.M.E1-Debeiky, N.E.Hasanien. The implementation o long-term forcasting strategies using a knowledge-based expert system Part-Ⅱ .Electric Power Systems Research (2001) 19-25
[44] T. Warren Liao,Z-H. Zhan,C.R. Mount. An integrated database and expert system for failure mechanism identification: Part Ⅱ-thesystem and performance testing. Enginerring Failure Analysis 6(1999)407-421
[45] K.Bryant. ALEES: an agricultural loan evaluation expert system. Expert System with Applications(2001) 75-85
[46] M.T. Elhadi .Bankruptcy support system: taking advantage of information retrieval and case-based reasoning. Expert Systems with Applications(2000) 215-219
[47] John H.Bradley, Richard D.Hausr, JR. A Framework for Expert System Implementation. Expert Systems Whti Applications,Vol.8,157-167,1995
[48] 程瑞琪．用C++开发专家系统的一种新方法、小型微型计算机系统，]999 Vol20No3
[49] 陈洁，方敏．基于事实的自动解释机制及其实现方法．合肥工业大学学报(自然科学版)，1999．8，vol．22．No4．
[50] 宛延闿，定海．面向对象分析和设计．清华大学出版社，2001．2
[51] 江明德．面向对象程序设计．电子工业出版社，1994
[52] 科德，尤尔登 邵维忠．面向对象的分析．北京大学出版社，1992．2
[53] 王道平，冯振声，董玉友．基于面向对象的专家系统设计方法研究．河北科技大学学报，Vol 20 No 2 1999
[54] 何家峰，叶虎年，李柱，面向对象的专家系统的研究．华中理工大学学报，Vol 27 No 8 1999
[55] 余明兴，吴明哲．Borland C++ Builder 5 实例精解．清华大学出版社，2001．6
[56] 同志工作室．C++Builder5.0数据库开发实例．人民邮电出版社，2001
[57] 刘滨．C++Builder 5 高级编程实例精解．国防工业出版社，2001
[2] 喻瑞祥．货币信用与银行．中国财政经济出版社，1980
[3] 吴晓灵，李德。金融业的风险管理与信用评估．中国金融出版社，1996
[4] 吴晶妹。中国企业与证券的资信评估．中信出版社，1994．6
[5] 吉晓辉，黄纪宪．国有商业银行的贷款管理．同济大学出版社，1993．12，108
[6] 财政部统计评价司．企业效绩评价问答．经济科学出版社，1999
[7] Lyn C.Thomas. A survey of credit and behavioural scoring: forecasting financial risk of lenging to consumers. International Journal of Forecasting(2000) 149-172
[8] William F. Treacy, Mark Carey, Federal Resserve Board. Credit risk rating systems at large US banks, Journal of Banking & Finance 24(2000), 167-201
[9] Michel Crouhy, Dan Galai,Robert Mark. Prototype risk rating system. Journal of Banking & Finance 25(2001)47-95
[10] A fuzzy credit-rating approach for commercial loans: a Taiwan case,Liang-Hsuan Chen,Tai-Wei Chiou,Department of Industrial Management Science,National Cheng Kung University, Tainan,Taiwan,ROC, the International Journal of Management Science(1999),407-419
[11] 叶武．基于人工神经网络的银行客户信用评价系统的研究．厦门大学自动化系硕士学位论文，2000
[12] Masanori Amano. Credit rationing of a Bayesian bank with simple screening technologies. Japan and the World Economy11(1999)545-556
[13] Erkki K.Laitinen, Prediting a corporate credit analyst's risk estimate by logistic and linear models.international Review of financial Analysis(1999) 97-121
[14] 张维，李玉霜，王春峰．递归分类树在信用风险分析中的应用．系统工程理论与实践，2000年第三期
[15] 王春峰，康莉．基于遗传规划方法的商业银行信用风险评估模型．系统工程理论与实践，2001年第二期
[16] 王春峰，万海晖，张维．基于神经网络技术的商业银行信用风险评估．系统工程理论与实践，1999．9
[17] David West. Neural network credit scoring models. Computers & Operations Research(2000) 1131-1152
[18] Romaniuk SG, Hall LO.Decisionmaking on credit-worthiness,using a fuzzy connectionist model.Fuzzy Sets and Systems 1992;48:15-22
[19] Michael B.Gordy. A comparative anatomy of credit risk models. Journal of Banking & Finance 24(2000) 119-149
[20] S.Kanungo, S.Sharma, P.K.Jain. Evaluation of a decision support system for credit management decisions. Decision Support Systems30 (2001) 419-436
[21] 王万森．人工智能原理及其应用．电子工业出版社，2000．9
[22] 蔡自兴，徐光佑．人工智能及其应用．清华大学出版社，1996
[23] 田盛丰，黄厚宽．人工智能与知识工程．中国铁道出版社，1998
[24] 尹朝庆，尹浩．人工智能与专家系统．中国水利水电出版社，2001
[25] 杨炳儒．知识工程与知识发现．冶金工业出版社，2000
[26] 曹文君．知识库系统原理及其应用．复旦大学出版社
[27] 施鸿宝，王秋荷．专家系统．西安交通大学出版社
[28] 程慧霞等．用C++建造专家系统．电子工业出版社，1996
[29] 石纯一等．基于解释的机器学习方法．清华大学出版社，广西科学技术出版社，1997
[30] 汪培庄，李洪兴．知识表示的数学理论．天津科学技术出版社，1994
[31] 刘叙华．基于归结方法的自动推理．科学出版社，1994
[32] 王耀南．智能控制系统 模糊逻辑专家系统神经网络控制．湖南大学出版社
[33] 郭志刚．社会统计分析方法——-SPSS软件应用．中国人民大学出版社，1999．12
[34] 卢纹岱，朱一力，沙捷，朱红兵．SPSS for Windows从入门到精通．电子工业出版社，1999
[35] W. Greene,Yasuhiro Sakai. Comments on "Sample selection in credit scoring models". Japan and the World Economy(1998) 317-320
[36] Differences of opinion and selection bias in the credit rating industry, Richard Cantor, Frank Packer ,Federal Reserve Bank of New York,33 Liberty Street, New York,NY 10045,USA, Journal of Banking & finance 21(1997) 1395-1417
[37] 王林树，鲍兰平，赵瑞清．概率论与数理统计．科学出版社，1999．8
[38] 陈雄华．基于神经网络和专家系统结合的企业信用评估研究．厦门大学自动化系硕士学位论文，2002．6
[39] 高严．如何阅读三张表．经济科学出版社，2000．10
[40] Chih-Chou Chiu, Ling-Jing Kao. Combining a neural network with a Rule-Based Expert System.Approach for short-term Power Load Forecasting in Taiwan.Expert Systems with Applications.vol 13 No.4(299-305) 1997
[41] Zhang Liuyuan Lin Chengde. A New Method for Credit Assessment in Commercial Bank. Proceedings of the International Conference on Modeling and simulation in distributed applications, 2001.9,804-808
[42] 钟佩思 高国安．基于神经网络模块的混合型方案设计知识库系统．机械设计 No．5，1995．5
[43] M.S.Kandil, S.M.E1-Debeiky, N.E.Hasanien. The implementation o long-term forcasting strategies using a knowledge-based expert system Part-Ⅱ .Electric Power Systems Research (2001) 19-25
[44] T. Warren Liao,Z-H. Zhan,C.R. Mount. An integrated database and expert system for failure mechanism identification: Part Ⅱ-thesystem and performance testing. Enginerring Failure Analysis 6(1999)407-421
[45] K.Bryant. ALEES: an agricultural loan evaluation expert system. Expert System with Applications(2001) 75-85
[46] M.T. Elhadi .Bankruptcy support system: taking advantage of information retrieval and case-based reasoning. Expert Systems with Applications(2000) 215-219
[47] John H.Bradley, Richard D.Hausr, JR. A Framework for Expert System Implementation. Expert Systems Whti Applications,Vol.8,157-167,1995
[48] 程瑞琪．用C++开发专家系统的一种新方法、小型微型计算机系统，]999 Vol20No3
[49] 陈洁，方敏．基于事实的自动解释机制及其实现方法．合肥工业大学学报(自然科学版)，1999．8，vol．22．No4．
[50] 宛延闿，定海．面向对象分析和设计．清华大学出版社，2001．2
[51] 江明德．面向对象程序设计．电子工业出版社，1994
[52] 科德，尤尔登 邵维忠．面向对象的分析．北京大学出版社，1992．2
[53] 王道平，冯振声，董玉友．基于面向对象的专家系统设计方法研究．河北科技大学学报，Vol 20 No 2 1999
[54] 何家峰，叶虎年，李柱，面向对象的专家系统的研究．华中理工大学学报，Vol 27 No 8 1999
[55] 余明兴，吴明哲．Borland C++ Builder 5 实例精解．清华大学出版社，2001．6
[56] 同志工作室．C++Builder5.0数据库开发实例．人民邮电出版社，2001
[57] 刘滨．C++Builder 5 高级编程实例精解．国防工业出版社，2001