A new adaptive neuro-fuzzy approach for on-line nonlinear system identification

A new on-line identification algorithm is presented in this paper based on a neuro-fuzzy model structure. The algorithm is developed based on the functional equivalence between a radial basis function (RBF) neural network and a fuzzy inference system (FIS). The developed algorithm utilizes a Weighted Rule Activation Record (WRAR) as a functional measure to monitor the modeling efficiency of the created rules. This measure evaluates the influence of each created rule with a time-based memory weight which puts more emphasis on the most recent input data. The proposed technique employs an extended Kalman filter (EKF) as a learning algorithm to adapt the antecedent and consequent parameters of the nearest rule. This algorithm benefits simple and understandable criteria to make it more attractive in practical applications. This leads to more efficient rule base with low created rules. Its low computational time makes it as an appropriate on-line identification approach. The performance of the proposed algorithm with some other new algorithms have been evaluated on a nonlinear dynamic system. Simulation results demonstrate the efficiencies of the proposed algorithm, resulting to the most simple rule structure with the lowest computational time.