A method for extracting a nonlinear time series prediction model by using a weighted fuzzy membership function based neural network is provided to enhance recognition rate and to improve reliability with respect to prediction of an input pattern because the number of fuzzy rules for pattern classification is small. A method for extracting a nonlinear time series prediction model comprises the following several steps. N property patterns are inputted into a weighted fuzzy membership function neural network(S201). A hyper box layer generates a hyper box node with n fuzzy sets with respect to the n property patterns(S202). The hyper box layer classifies class nodes by calculating average values of membership functions affiliated to the hyper box node via an intensified node output equation(S203). Then, membership functions with new fixed points and weighting values are obtained by adjusting the membership functions and their weighting values of the hyper box node via the weighted fuzzy membership function based neural network learning algorithm(S204). Fuzzy rules are extracted by integrating fuzzy properties of the membership functions with the new fixed points and weighting values(S205).
展开▼
