基于BP神经网络的企业信用评级模型  

Enterprise credit rating model based on BP neural networks

作　　者：张鸿[1] 丁以中[1] 

机构地区：[1]上海海事大学经济管理学院,上海200135

出　　处：《上海海事大学学报》2007年第3期64-68,共5页Journal of Shanghai Maritime University

基　　金：上海市教委重点项目(06ZZ43)

摘　　要：通过科学的方法对企业信用进行分析、评级和判断,给出定性与定量相结合的指标体系,建立基于BP神经网络的两类企业信用评级模型.针对局部收敛的缺点,用自适应学习率和附加动量项改进信用评级,并运用该模型对我国2004年100家ST和非ST上市公司进行评级,得出对训练样本和测试样本的评级准确率,表明神经网络技术作为智能化科学方法,非常适合企业信用评级,但也存在网络稳定性差等不足.To analyze, evaluate and judge the credit of enterprises with scientific methods, a credit rating index system is proposed. The credit rating model is established for two kinds of enterprises based on BP neural networks. With the local convergence defect, the self-adaptive learning rate and additional item of momentum are used to improve credit rating. The model is applied in credit rating of one hundred ST and none-ST listed companies in 2004. The accuracy on credit rating is obtained for training samples and test samples. It is shown that, as a scientific intelligent method, BP neural network is very suitable for credit rating of enterprises, but it also has defects such as poor network stability.

关 键 词：BP神经网络 信用评级 指标体系 自适应学习率 动量项 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程] F830.39[自动化与计算机技术—控制科学与工程]