A Research Of Fuzzy Neural Network Model Based On Rules Reduction

Fuzzy algorithm and fuzzy logic into neural network become the so-called fuzzy neuralnetwork.Fuzzy logic method and the neural network method together is the biggestcharacteristic of fuzzy neural network, the defects of the neural network in fuzzy dataprocessing deficiencies and pure fuzzy logic in learning has been improved due to the fusionof both. At the same time, the fuzzy neural network’s input-output relationship can realizemany to be able to use the rules state causality, adaptive can be learned to achieve thefunction,and precise or vague expression knowledge can automatic gain.The various problems currently widely applied to the fuzzy neural network, but thecommonly used fuzzy neural network model has n inputs,each input to take m membershipfunctions, rules the number of layers for the collection of all the rules of the mn,so that wouldmake the structure of the fuzzy neural network becomes huge become possible. Slow learningspeed is a huge structure to bring a direct problem,and the structure is a huge limiting theapplication of fuzzy neural network in multi-input fuzzy system. Complete structure afterlearning of fuzzy neural network, although the network structurewe can through the clip rulesto reduce,but only fuzzy neural network in the later application is limited to reducecomputation; The huge structure of the previous study still need to use training,a lot of timeand hardware resources still need to spend away, at the same time for the hardwarerequirements also increased, Therefore, this subject studies this issues, in order to improve thefuzzy neural network approximation and efficiency performance.This paper reduces the rules of fuzzy neural network model through subtractiveclustering.include:1) use of Subtractive clustering,can continue to choose the most densethrough constant center and the radius of the right,according to the data center classified,sothat the data gathered in the vicinity of the cluster center.2) after Subtractiveclustering,category corresponds to the center of the data samples can be extracted,and theform of fuzzy rules that these centers.When any data input,will distinguish the data andcategories of affiliation to the possibility of discrimination rules corresponding to the resultoutput,and choose the output likelihood of the results.Not only to achieve a reduction of fuzzyrules,and also accelerate the network convergence speed and reduce the complexity of themodel structure.3) With the software Matlab,this paper compares subtractive clusteringbased on fuzzy neural network model with the general fuzzy neural network model,the resultof the got90%reduction in fuzzy rules under the conditions of high recognition rate right of fuzzy neural network.Further,the application based on rules reduction of fuzzy neural networkmodel for enterprise electronic commerce development level comprehensive evaluation,andobtains satisfactory results.