A neuro-fuzzy model for nonlinear plants identification
Ieroham Solomon Baruch, Elena Gortcheva, Federico Thomas, Ruben A. Garrido · DIGITAL.CSIC (Spanish National Research Council (CSIC)) · 1999
A improved parallel Recurrent Neural Network (RNN) model and an improved dynamic Backpropagation (BP) method of its learning, are proposed.The RNN model is given as a two layer Jordan canonical architecture for both continuous and discrete-time cases.The output layer is of Feedforward type.The hidden layer is a recurrent one with self-feedbacks and full forward connections with the inputs.A linearisation of this RNN model is performed and the stability, observability and controllability conditions, are studied.To preserve the RNN stability, sigmoid activation functions are introduced in RNN feedback loops.The paper suggests to improve RNN realisation using saturation function instead of a sigmoid one.A new improved RNN learning algorithm of dynamic BP-type containing momentum term, is proposed.For a complex non-linear plants identification, a fuzzyrule-based system and a neuro-fuzzy model, are proposed.The proposed neuro-fuzzy model is applied for identification of a mechanical system with friction.