基于贝叶斯正则化神经网络的企业资信评估  

Bayesian - Regularization Neural Network for Corporation Credit Rating

作　　者：罗烨[1] 柳益君[1] 朱广萍[1] 叶飞跃[1] 

机构地区：[1]江苏技术师范学院计算机科学与工程学院,江苏常州213001

出　　处：《计算机仿真》2010年第11期303-306,共4页Computer Simulation

基　　金：常州市"831工程"资助项目(KYZ06002);江苏技术师范学院青年基金项目(KYY06075)

摘　　要：在市场经济系统研究中,资信评估作为市场经济中的监督力量,是投资者的重要参考依据。科学准确的资信评估可以辅助决策,降低投资者风险。针对当前企业资信评估方法的不足,为了准确评估资信效果,提出将基于贝叶斯正则化的前向多层神经网络用于企业资信评估,通过新的误差函数可以减少网络的有效权值和阈值,并使网络训练输出更加平滑,从而增强网络的泛化性能。并通过MATLAB软件及其神经网络工具进行仿真计算。结果表明,贝叶斯正则化神经网络稳定、快捷、评价结果可靠准确,可作为于企业资信评估依据。As the monitoring forces in market economy,credit rating is an important reference for investors.Accurate and scientific credit rating can help decision-making and reduce the risk of investors.To avoid the inability of conventional methods of corporate credit rating,the paper proposes to apply the feed forward artificial neural network with Bayesian-regularization training algorithm to the problem of credit rating.The adoption of a new error function can reduce the network effective weights and thresholds and make the output of network training smoother,so that the network generalization performance is enhanced.Matlab software and its neural network toolbox are used to simulate and compute.The experiment results show that the Bayesian-regularization neural network is stable and fast,has good prediction accuracy,and can be applied to the corporate credit rating.

关 键 词：神经网络 贝叶斯正则化 资信评估 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]