银行信贷风险评价及预警RBFNN系统设计研究
|
摘要
控制银行贷款风险一直是我国商业银行经营的重中之重。本文针对传统方法有相应的缺陷,提出了一个新的评价银行贷款风险的方法——基于均匀设计法的RBFNN风险评价及预警评价方法。同时本文运用面向对象分析和面向对象设计方法对银行贷款风险评价及预警系统进行了设计。其主要工作其主要分成以下几个阶段:
1、在分析银行贷款评价和预警评价的流程基础上,对项目进行了可行性分析,接着运用OOA方法进行了需求分析,进行了详细的各层次用例分析。
2、在选取了优秀的银行贷款风险评价及预警评价指标基础上,运用所设计的指标模糊隶属度函数对评价指标值进行规范化,并使用均匀设计法获取了自适应RBFNN训练样本。通过样本训练和实例数据进行检验贷款风险评价和预警评价,获得了理想的结果。
3、运用UML建模,对银行贷款风险评价及预警系统进行了总体设计和详细设计,设计出类图、系统模块结构图,并对各模块设计出较详细的时序图。
4、在系统实现过程中,运用了Matlab函数与C#混编技术,将Matlab的自适应RBFNN函数转换成COM组件,在C#程序中调用。同时应用了均匀设计法所获得的大量样本,分别进行贷款风险评价的RBFNN训练和预警评价的RBFNN训练,经过程序检验,模型训练检验满意。贷款风险评价和预警评价,也获得理想的结果。从而使应用系统脱离了Matlab软件而运行,也简化了RBFNN算法的编写。
Matlab函数与C#混编技术的运用,使我们今后在C#程序中可以运用Matlab软件的大量的函数进行复杂计算和图形显示变换,简化了程序编制,同时充分发挥Matlab软件强大的计算功能。
在过去的银行风险预警评价方法中,传统模糊综合评判法具有无法摆脱决策过程中的随机性,主观不确定性以及缺乏与时俱进的自学习能力的严重缺点。过去采用的RBF神经网络评价方法具有网络隐节点固定,局部极值的缺点。虽然它具有自学习能力的优点,但是在自学习训练中却具有致命的缺点——在有限的学习样本情况下缺乏严密的实验设计,从而使所训练出的网络性能难以覆盖绝大多数风险情况,致使人工神经网络评价性能不能达到预期要求。
本系统利用“均匀设计法”U1000表,通过DSP软件,获取大规模标准化“均匀设计”样本,然后利用这些能够覆盖全部样本空间的有代表性样本进行自适应RBFNN训练,训练检验获取了误差小于预定误差的模型训练结果;实例检验获得了误差小于预定误差绝对值3.5,预警等级符合实际的应用效果。从而建立一套较准确的、正确的、无遗漏的、快速的银行贷款风险评价及预警评价模型和评价方法。
本系统的设计为以上方法奠定了实际运用的平台,对于银行贷款评价和预警评价具有实际运用价值。
It is always the most important for China's commercial banks to control loaning risk. According to limitation of traditional loaning risk assessment, this paper proposed a new evaluation of the risk of bank loans - based on even distributed combining of the RBFNN risk assessment and early risk warning assessment evaluation methods. At the same time, it use the object-oriented analysis and object-oriented design methods of bank loans risk assessment and early warning systems for the design. The main divided into the following phases:
1. First of all, the analysis of bank loans evaluation and early warning based on the evaluation of the process of the project feasibility analysis, and then use OOA methods for require analysis, carried out a detailed analysis of all levels use cases.
2. With outstanding bank loans risk assessment and early warning indicators on the basis of evaluation. The corporate lending risk evaluation index value and credit risk early warning indicators of value standards, the use of uniform design algorithm RBFNN access to the index system. Through examples of data for risk evaluation and warning evaluation, get the desired results.
3. Use UML modeling, the bank loan risk assessment and early warning systems for the system design and detailed design, design a class diagram, the module structure of the system. And the modular design to a more detailed Sequence Diagram.
4. In the system to achieve process, the use of Matlab functions and C # mixed technology, build Matlab Adaptive RBFNN function into COM components, in the process C # program to calling. At the same time application of a uniform design given the large number of samples were carried out risk assessment of the loan RBFNN training and evaluation of the early warning RBFNN training, after testing procedures, test satisfied with the training model. Loan evaluation of risk assessment and early warning, get the desired results. So that the program can run without Matlab software, and simplify the RBFNN code.
With Matlab functions and C # mixed program so that our future in C # procedures can be used in Matlab software function of the large number of complex calculations and graphics transform and simplify the procedures for the preparation, at the same time giving full play to Matlab software powerful computing capabilities.
In the past the bank risk early warning evaluation methods, the traditional fuzzy comprehensive evaluation method is not out of the decision-making process of random, subjective uncertainty and lack of self-learning ability to advance with the times of serious shortcomings. Used in the past RBF neural network evaluation method is implicit node fixed network, local extreme deficiencies. Although it has the advantages of self-learning ability, but the self-study training is a fatal drawback - in limited circumstances study samples of the lack of rigorous experimental design, so that by training hard to cover the vast majority of network performance risks, Evaluation of artificial neural network which can not achieve the expected performance requirements.
We use the "uniform design" U1000 table, through the DSP software, access to large-scale standardized "uniform design" sample, and then use these to cover all of a representative sample space samples RBFNN adaptive training, training access to the test Error is less than the expected error model training results; examples of the test was scheduled less than the absolute value of 3.5, early warning levels in line with the actual effect. So as to establish a more accurate and correct, without missing, rapid bank lending risk assessment and early warning assessment model and evaluation methods.
The design of the system laid the way for more than a platform for practical use, bank loans evaluation and early warning assessment of the value of practical use.
1、在分析银行贷款评价和预警评价的流程基础上,对项目进行了可行性分析,接着运用OOA方法进行了需求分析,进行了详细的各层次用例分析。
2、在选取了优秀的银行贷款风险评价及预警评价指标基础上,运用所设计的指标模糊隶属度函数对评价指标值进行规范化,并使用均匀设计法获取了自适应RBFNN训练样本。通过样本训练和实例数据进行检验贷款风险评价和预警评价,获得了理想的结果。
3、运用UML建模,对银行贷款风险评价及预警系统进行了总体设计和详细设计,设计出类图、系统模块结构图,并对各模块设计出较详细的时序图。
4、在系统实现过程中,运用了Matlab函数与C#混编技术,将Matlab的自适应RBFNN函数转换成COM组件,在C#程序中调用。同时应用了均匀设计法所获得的大量样本,分别进行贷款风险评价的RBFNN训练和预警评价的RBFNN训练,经过程序检验,模型训练检验满意。贷款风险评价和预警评价,也获得理想的结果。从而使应用系统脱离了Matlab软件而运行,也简化了RBFNN算法的编写。
Matlab函数与C#混编技术的运用,使我们今后在C#程序中可以运用Matlab软件的大量的函数进行复杂计算和图形显示变换,简化了程序编制,同时充分发挥Matlab软件强大的计算功能。
在过去的银行风险预警评价方法中,传统模糊综合评判法具有无法摆脱决策过程中的随机性,主观不确定性以及缺乏与时俱进的自学习能力的严重缺点。过去采用的RBF神经网络评价方法具有网络隐节点固定,局部极值的缺点。虽然它具有自学习能力的优点,但是在自学习训练中却具有致命的缺点——在有限的学习样本情况下缺乏严密的实验设计,从而使所训练出的网络性能难以覆盖绝大多数风险情况,致使人工神经网络评价性能不能达到预期要求。
本系统利用“均匀设计法”U1000表,通过DSP软件,获取大规模标准化“均匀设计”样本,然后利用这些能够覆盖全部样本空间的有代表性样本进行自适应RBFNN训练,训练检验获取了误差小于预定误差的模型训练结果;实例检验获得了误差小于预定误差绝对值3.5,预警等级符合实际的应用效果。从而建立一套较准确的、正确的、无遗漏的、快速的银行贷款风险评价及预警评价模型和评价方法。
本系统的设计为以上方法奠定了实际运用的平台,对于银行贷款评价和预警评价具有实际运用价值。
It is always the most important for China's commercial banks to control loaning risk. According to limitation of traditional loaning risk assessment, this paper proposed a new evaluation of the risk of bank loans - based on even distributed combining of the RBFNN risk assessment and early risk warning assessment evaluation methods. At the same time, it use the object-oriented analysis and object-oriented design methods of bank loans risk assessment and early warning systems for the design. The main divided into the following phases:
1. First of all, the analysis of bank loans evaluation and early warning based on the evaluation of the process of the project feasibility analysis, and then use OOA methods for require analysis, carried out a detailed analysis of all levels use cases.
2. With outstanding bank loans risk assessment and early warning indicators on the basis of evaluation. The corporate lending risk evaluation index value and credit risk early warning indicators of value standards, the use of uniform design algorithm RBFNN access to the index system. Through examples of data for risk evaluation and warning evaluation, get the desired results.
3. Use UML modeling, the bank loan risk assessment and early warning systems for the system design and detailed design, design a class diagram, the module structure of the system. And the modular design to a more detailed Sequence Diagram.
4. In the system to achieve process, the use of Matlab functions and C # mixed technology, build Matlab Adaptive RBFNN function into COM components, in the process C # program to calling. At the same time application of a uniform design given the large number of samples were carried out risk assessment of the loan RBFNN training and evaluation of the early warning RBFNN training, after testing procedures, test satisfied with the training model. Loan evaluation of risk assessment and early warning, get the desired results. So that the program can run without Matlab software, and simplify the RBFNN code.
With Matlab functions and C # mixed program so that our future in C # procedures can be used in Matlab software function of the large number of complex calculations and graphics transform and simplify the procedures for the preparation, at the same time giving full play to Matlab software powerful computing capabilities.
In the past the bank risk early warning evaluation methods, the traditional fuzzy comprehensive evaluation method is not out of the decision-making process of random, subjective uncertainty and lack of self-learning ability to advance with the times of serious shortcomings. Used in the past RBF neural network evaluation method is implicit node fixed network, local extreme deficiencies. Although it has the advantages of self-learning ability, but the self-study training is a fatal drawback - in limited circumstances study samples of the lack of rigorous experimental design, so that by training hard to cover the vast majority of network performance risks, Evaluation of artificial neural network which can not achieve the expected performance requirements.
We use the "uniform design" U1000 table, through the DSP software, access to large-scale standardized "uniform design" sample, and then use these to cover all of a representative sample space samples RBFNN adaptive training, training access to the test Error is less than the expected error model training results; examples of the test was scheduled less than the absolute value of 3.5, early warning levels in line with the actual effect. So as to establish a more accurate and correct, without missing, rapid bank lending risk assessment and early warning assessment model and evaluation methods.
The design of the system laid the way for more than a platform for practical use, bank loans evaluation and early warning assessment of the value of practical use.
引文
[1]巴塞尔银行监管委员会文献汇编.中国金融出版社,2002(9):15-17
[2]王春峰.金融市场风险管理.中国人民大学出版社,2002(1)
[3]张皓宇.银行信贷风险的成因、预警和化解.商业研究. 2003(14)
[4]马兰芳,刘金兰,杨军等.人工神经网络在商业银行监测预警中的应用研究管理工程学报,2002.2:5-8
[5]杨保安,朱明.基于神经网络与专家系统结合的银行贷款风险管理.系统工程理论方法应用, 1999.8:7-9
[6] Kamel M. MODULAR NEURAL NETWORKS: A SURVEY. International Journal of Neural Systems (IJNS) , 1999,9():129-151
[7]杨保安,季海.基于人工神经网络的商业银行贷款风险预警研究.系统工程理论与实践,2001.5:70-74.
[8] Joseph Giarratano, Gary Riley. Expert Systems Principles and Programming. PWS Pub, Co. 2002.8
[9] OMG Unified Modeling Language Specification. Object Management Group, Framingham, MA, 1998. Internet:www.omg.org
[10]吕江林.投资银行风险预警机制初探[J].金融与经济, 2001.1:37- 39.
[11]方开泰.王元.数论方法在统计中的应用[M].北京:科学出版社,1996.
[12]方开泰.均匀设计和均匀设计表[M].科学出版社,1994.6
[13] Simon Robinson. C#高级编程.清华大学出版社,2005.6
[14]闻新,周露,李翔等. Matlab神经网络仿真与应用.科学出版社,2003.7
[15]高隽.人工神经网络原理及仿真实例.机械工业出版社,2003
[16] Fredric M. Ham, Ivica Kostanic. Principles of Neurocomputing for Science & Engineering. PWS Pub, Co. 2003.7
[17] Martin T, Hagan Howard B, Demuth. Neural Network Design. PWS Pub, Co. 2002.8
[18] H M Sneed. Encapsulating legacy software for use in client/serversystems, Proc. IEEE Conference on Software Maintenance, 1996:104-120
[19]吴晓灵、李德.金融业的风险管理与信用评估[M].北京,中国金融出版社,1996
[20] P Kruchten. Architectural blueprints-the“4+1”view model of software architecture. IEEE Software, 1995(6):42~50
[21] Karl Wieger. "Use cases: Listening to the customer's voice,". Software Development, 1997.3
[22]贺晓波,张宇红.我国商业银行风险预警指标体系的构建.管理现代化,2001(4)
[23]赵国杰.银行项目贷款决策流程与风险评估体系的建构.管理工程学报,2003.1:98-101
[24]陈军.关于信贷风险预警的研究.上海金融,2000(11)
[25]谭中明.运用层次分析法进行银行信贷决策.江苏理工大学学报,1995
[26]黄祖勤.商业银行信贷管理.广东经济出版社,1999
[27]薛锋,乔卓.神经网络模型在上市公司财务困境预测中的应用[J].西安交通大学学报, 2003.23(2):22- 25.
[28]谢赤.金融风险管理中的最优化模型[J].预测1996.6:45-48
[29]张川,佟玉明,潘德惠.商业银行经营风险评价指标体系及模糊综合评判.东北大学学报,2003.11
[30] Simon Haykin.神经网络原理(原书第2版).机械工业出版社,2004.5
[31] http://www.cs.stir.ac.uk/~lss/NNIntro/InvSlides.html Monday, 24-Jul-2006 09:23:17 BST
[32]孙延风,梁艳春.改进的神经网络最近邻聚类学习算法及其应用.吉林大学学报. 2002(3)35-37
[33] Andrea Beltratti, Margarita P. Terna. Neural Networks for Economic and Financial Modelling.International Thomson Computer Press. 1995.12
[34]包约翰,马颂德等.自适应模式识别与神经网络.科学出版社,1992
[35]张美恋,王秀珍.基于径向基神经网络的商业银行风险预警系统研究.集美大学学报,2000.3:280-284
[36]方开泰.均匀设计及其应用(Ⅰ)[J].数据统计与管理,1994.13:57-63.
[37]徐晋.基于动态神经网络专家系统的供应商信用等级分析[J].软科学,北京: 2003.17: 5- 8
[38]余力,许兆龙等.商业银行风险的防范与规避.西安交通大学出版社,1996
[39]郝丽萍、胡欣悦、李丽.商业银行信贷风险分析的人工神经网络模型.系统工程理论与实践,2001.21(5):62-71.
[40] Alexander J A, Mozer M C. Template-based procedures for neural network interpretation. Neural Networks, 1999.12(3):479-498
[41] Amato J D, Thomas L. Forecase-based monetary policy[J]. FRB Kansas City, RW, October, 1999,99-100
[42] Flanagan , R and Norman. G Risk Management and Constructions Blackwell Scientific. U K,1993.1
[43] N Medvidovic, et al. Modeling software architectures in the unified modeling language.http://Sunset.usc.edu/csse/TECHRPTS/2000/usccse2000-512/usccse2000-512.pdf,2000:1-57
[44] Ron Patton.软件测试.周予滨,姚静等.北京. 2002.2
[2]王春峰.金融市场风险管理.中国人民大学出版社,2002(1)
[3]张皓宇.银行信贷风险的成因、预警和化解.商业研究. 2003(14)
[4]马兰芳,刘金兰,杨军等.人工神经网络在商业银行监测预警中的应用研究管理工程学报,2002.2:5-8
[5]杨保安,朱明.基于神经网络与专家系统结合的银行贷款风险管理.系统工程理论方法应用, 1999.8:7-9
[6] Kamel M. MODULAR NEURAL NETWORKS: A SURVEY. International Journal of Neural Systems (IJNS) , 1999,9():129-151
[7]杨保安,季海.基于人工神经网络的商业银行贷款风险预警研究.系统工程理论与实践,2001.5:70-74.
[8] Joseph Giarratano, Gary Riley. Expert Systems Principles and Programming. PWS Pub, Co. 2002.8
[9] OMG Unified Modeling Language Specification. Object Management Group, Framingham, MA, 1998. Internet:www.omg.org
[10]吕江林.投资银行风险预警机制初探[J].金融与经济, 2001.1:37- 39.
[11]方开泰.王元.数论方法在统计中的应用[M].北京:科学出版社,1996.
[12]方开泰.均匀设计和均匀设计表[M].科学出版社,1994.6
[13] Simon Robinson. C#高级编程.清华大学出版社,2005.6
[14]闻新,周露,李翔等. Matlab神经网络仿真与应用.科学出版社,2003.7
[15]高隽.人工神经网络原理及仿真实例.机械工业出版社,2003
[16] Fredric M. Ham, Ivica Kostanic. Principles of Neurocomputing for Science & Engineering. PWS Pub, Co. 2003.7
[17] Martin T, Hagan Howard B, Demuth. Neural Network Design. PWS Pub, Co. 2002.8
[18] H M Sneed. Encapsulating legacy software for use in client/serversystems, Proc. IEEE Conference on Software Maintenance, 1996:104-120
[19]吴晓灵、李德.金融业的风险管理与信用评估[M].北京,中国金融出版社,1996
[20] P Kruchten. Architectural blueprints-the“4+1”view model of software architecture. IEEE Software, 1995(6):42~50
[21] Karl Wieger. "Use cases: Listening to the customer's voice,". Software Development, 1997.3
[22]贺晓波,张宇红.我国商业银行风险预警指标体系的构建.管理现代化,2001(4)
[23]赵国杰.银行项目贷款决策流程与风险评估体系的建构.管理工程学报,2003.1:98-101
[24]陈军.关于信贷风险预警的研究.上海金融,2000(11)
[25]谭中明.运用层次分析法进行银行信贷决策.江苏理工大学学报,1995
[26]黄祖勤.商业银行信贷管理.广东经济出版社,1999
[27]薛锋,乔卓.神经网络模型在上市公司财务困境预测中的应用[J].西安交通大学学报, 2003.23(2):22- 25.
[28]谢赤.金融风险管理中的最优化模型[J].预测1996.6:45-48
[29]张川,佟玉明,潘德惠.商业银行经营风险评价指标体系及模糊综合评判.东北大学学报,2003.11
[30] Simon Haykin.神经网络原理(原书第2版).机械工业出版社,2004.5
[31] http://www.cs.stir.ac.uk/~lss/NNIntro/InvSlides.html Monday, 24-Jul-2006 09:23:17 BST
[32]孙延风,梁艳春.改进的神经网络最近邻聚类学习算法及其应用.吉林大学学报. 2002(3)35-37
[33] Andrea Beltratti, Margarita P. Terna. Neural Networks for Economic and Financial Modelling.International Thomson Computer Press. 1995.12
[34]包约翰,马颂德等.自适应模式识别与神经网络.科学出版社,1992
[35]张美恋,王秀珍.基于径向基神经网络的商业银行风险预警系统研究.集美大学学报,2000.3:280-284
[36]方开泰.均匀设计及其应用(Ⅰ)[J].数据统计与管理,1994.13:57-63.
[37]徐晋.基于动态神经网络专家系统的供应商信用等级分析[J].软科学,北京: 2003.17: 5- 8
[38]余力,许兆龙等.商业银行风险的防范与规避.西安交通大学出版社,1996
[39]郝丽萍、胡欣悦、李丽.商业银行信贷风险分析的人工神经网络模型.系统工程理论与实践,2001.21(5):62-71.
[40] Alexander J A, Mozer M C. Template-based procedures for neural network interpretation. Neural Networks, 1999.12(3):479-498
[41] Amato J D, Thomas L. Forecase-based monetary policy[J]. FRB Kansas City, RW, October, 1999,99-100
[42] Flanagan , R and Norman. G Risk Management and Constructions Blackwell Scientific. U K,1993.1
[43] N Medvidovic, et al. Modeling software architectures in the unified modeling language.http://Sunset.usc.edu/csse/TECHRPTS/2000/usccse2000-512/usccse2000-512.pdf,2000:1-57
[44] Ron Patton.软件测试.周予滨,姚静等.北京. 2002.2