Dynamic Behavior recurrent neuro-fuzzy modeling by combining global learning and local learning

To improve the interpretability and generalization of recurrent neuro-fuzzy model, we propose a two-phase training scheme. In the phase of local learning, we tune local models using Fuzzy C-Means clustering method and Recursive Least Squares technique. In the phase of global learning, we refine the recurrent neuro-fuzzy model using the BP algorithms. In the end of paper, we illustrate the validity and efficiency of algorithms using a stoker-fired boiler example.