Neural Network Based On The Fuzzy Random Neural Network Model And Application

Fuzzy neural network was developed based on the neural network and fuzzy system.It sets language computing,logic,distributed processing and nonlinear dynamic process as a whole. Although fuzzy neural network has been widely used, but there are still a lot of room for improvement. Currently the correction algorithm parameters currently used more widely in the training process of fuzzy neural network is the inverse error correction algorithm, which is easy to fall into local minimum and slow convergence; When the combination of variable fuzzy division increased,it will lead to "explosion"; Most of the current fuzzy neural network only model for fuzzy variables and fuzzy random variables simultaneously contain little involved.This paper studied the problems above and proposes a new neural network which is based on fuzzy random variables.Including:(1) In this paper,we use subtractive clustering to extract rules.Subtractive Clustering can be based on the sample information and automatically access rules which can achieve the limited coverage of rules.(2) In this paper we proposed the overall error rate adaptive learning algorithm of the parameters corrected.In the learning process,the global error adaptive learning rate adjustment in accordance with the training process can reduce errors in turmoil and the error can faster convergence to a minimum.(3) There is mixed uncertainty variables in practical application modeling,but most of the current fuzzy neural network only modeling for fuzzy variables and rarely contain both.In this paper,based on fuzzy neural network which is on the basis of subtractive clustering we proposed the fuzzy random neural network.By converting the output variable to get samples of different fuzzy partition.The output value of the variable level as a liner combination of weights to get integrated variables.Finally according to the distribution to get the prediction interval.Last,we use the data of the underground gas safety grade evaluation to model the random fuzzy neural network.Comparing the results of random fuzzy neural network and the results of fuzzy neural network which is based on the subtraction clustering.On the basis of MATLAB software support,the simulation results show that the results of the model we put forward improve a lot when comparing with the fuzzy neural network which is based on subtractive clustering in results and realize the rules subtract so it is of stronger explanatory.