An adaptive neural fuzzy filter and its applications

摘要: A new kind of nonlinear adaptive filter, the neural fuzzy filter (ANFF), based upon a network's learning ability and if-then rule structure, is proposed in this paper. The ANFF inherently feedforward multilayered connectionist network which can learn by itself according to numerical training data or expert knowledge represented rules. adaptation here includes construction rules (structure learning), tuning free parameters membership functions (parameter learning). In structure phase, are found on matching input-output clusters. parameter backpropagation-like algorithm developed minimize output error. There no hidden nodes (i.e., rules) initially, both performed concurrently as proceeds. However, if some linguistic information about design available, such be put into form an initial with nodes. Two major advantages thus seen: 1) priori incorporated makes fusion possible; 2) predetermination, like number nodes, must given, since find its optimal automatically.