信用风险预警的MA-RSFNN模型构建与应用

A Fuzzy Neural Network based on Memetic Algorithms and Rough Set Theory for Credit Risk Early Warning and Its Application

为了克服模糊神经网络的维数灾难、结构复杂、局部早熟及收敛慢等缺陷,在设计一种模糊神经网络的基础上,将模因算法和粗糙集理论引入模糊神经网络,提出一种模因进化型粗糙模糊神经网络(MA-RSFNN)。新模型借助模因算法的全局搜索能力减少网络陷入局部极值的可能性,同时利用粗糙集知识约简对网络输入数据进行降维消冗,精简输入维度,避免"维数灾难"。实例仿真结果表明MA-RSFNN模型的预测准确性较高,是一类解决金融风险管理中高维复杂问题的有效方法。

In order to overcome the drawbacks of fuzzy neural network such as curse of dimensionality, complex structure, local optimization, slow convergence and so on, a memetic optimized rough fuzzy neural network is proposed by integrating the Memetic Algorithms（MA） and Rough Set（RS） into Fuzzy Neural Network（FNN） after that a new FNN has been de- signed. The new model can reduce the possibility of network into local extremes by taking advantage of the MA＇ s global search ability, and can avoid the curse of dimensionality by using the knowledge reduction of rough set to reduce dimension- ality and eliminate redundancy of the input data set. The application result indicates that the MA - RSFNN could obtain much higher accuracy of prediction. It is an. efficient＂ method to solve high dimensional complex problems of financial risk management.