基于模糊粗糙模型的粗神经网络建模方法研究

Fuzzy Rough Model Based Rough Neural Network Modeling

提出一种基于模糊粗糙模型的粗神经网络建模(FRM_RNN_M)方法.该方法通过自适应G-K聚类实现输入输出积空间的模糊划分,进而在聚类数和约简属性搜索的基础上,提取优化的模糊粗糙模型(Fuzzy rough model,FRM),并在融合神经网络后实现粗神经网络建模.分类实验表明,FRM_RNN_M的分类性能优于传统贝叶斯和LVQ方法,而且比单纯的FRM模型具有更强的综合决策能力,和传统的粗逻辑神经网络(Rough logic neural network,RLNN)相比,FRM_RNN_M方法建立的神经网络结构精简,收敛速度快,具有更强的泛化能力.

The rough neural network model based on fuzzy rough model （FRM）, FRM_RNN.M, is addressed. By means of adaptive Gaustafason Kessel （G-K） algorithm, fuzzy partition can be implemented in the input-output product space. Based on the search of cluster number and feature reduction, optimized FRM can be extracted and rough neural network model can be constructed after integrating the neural network technique. Experiment results indicate that FRM_RNN_M is superior to conventional Bayesian and LVQ methods. Moreover, it has a better synthesis decision-making ability than the single FRM. Compared with conventional rough logic neural network （RLNN）, the neural network based on FRM_RNN_M has superiorities in size of structure, convergence speed, and generalization ability.