Nonlinear systems parameters estimation using radial basis function network

In this paper, a new on-line scheme for the state and parameter estimation of a large class of nonlinear systems is presented. This scheme uses a radial basis function neuronal predictor with the on-line learning of weights. The algorithms developed are potentially useful for adjusting the controller parameters of variable speed drives. The other interesting feature of the proposed method is its application to failure and fault detection. The parameter identification scheme is an algebraic method combined with state estimation. The asymptotic convergence of the estimates to their nominal values is achieved using the Lyapunov's arguments. The simulation results and the real-time estimation of both rotor resistance and speed of an induction motor based on this approach, show rapidly converging estimates in spite of the measurements noise, discretization effects, parameters uncertainties (e.g. inaccuracies on motor inductance values) and modeling inaccuracies. The other applications of the proposed method include the on-line estimation of the parameters of a synchronous generator.