RON研究及在信用风险评估中的应用

Research resource-optimizing network and its application in credit rating

针对资源分配网络算法鲁棒性较差,泛化能力不能保证等缺点,提出了一种将滑动数据窗口技术和梯度法相结合的资源优化网络,同时为避免网络过于庞杂,实现了隐节点的合并.在此基础上,建立了一种基于混杂神经网络的信用评估模型,并用该评估模型对上市公司的信用评级,表明了该评估模型的可行性和有效性.

To overcome disadvantages of Resource-Assigning Network algorithm such as bad robustness and weak generalization, Resource-Optimizing Network is proposed, which is combined sliding data window with grads. Meanwhile, to avoid numerous and jumbled network, this algorithm achieves combination of hidden neurons. On the basis, a credit rating model based on mixed neural network is proposed and applied to lots of listed companies. The result shows the feasibility and effectiveness of the evaluation model.