A NEW ADAPTIVE FUZZY INFERENCE NEURAL NETWORK

When the number of input dimension is large, the conventional fuzzy neural systems often cannot handle the task correctly because the degree of each rule becomes too small. In this paper, a new adaptive fuzzy inference neural network (AFINN) based on full implication triple-I fuzzy inference method has been described. It has fuzzy weights and accepts fuzzy set inputs. The advantages of AFINN are that it has initial rule creation ability and fuzzy inference ability, the degree of each rule doesn't need to be calculated. This system also automatically generates rules from numerical data. The proposed system operates with Gaussian membership functions in premise and conclusion parts. Euclidian distances are used to parameter estimation and initialization of unknown parameter values. For evaluation of the number of if-then rules, the standard RMSE performance index have been applied. The applications to prediction of chaotic time series is considered in this paper as well.