An Improved Feedforward Fuzzy Neural Network and Its Learning Algorithm

摘要

Symmetric polygonal fuzzy number is employed to construct an improved feedforward fuzzy neural network(FNN). First, a novel fuzzy arithmetic and extension principle for such polygonal fuzzy numbers is derived. Second, the topological architecture of a three layer feedforward FNN is presented, and the input-output law of such a network is systematically studied. Third, a fuzzy BP learning algorithm for the polygonal fuzzy number connection weights and thresholds of the FNN is developed. Finally a simulation example is analyzed to realize approximately data pairs whose values are real numbers and symmetric polygonal fuzzy numbers, by the adaptive three layer feedforward FNN.