Adaptive step size kernel least mean square algorithm for Lorenz time series prediction

An adaptive stepsize kernel least mean square(AKLMS) algorithm is presented in this paper. A mechanism is introduced here to adjust the stepsize parameter in the output of the KLMS algorithm using gradient descent method. The proposed method improves the results by reducing the training time of the algorithm that also helps in converging to global minima instead of local minima. We named the algorithm, the adaptive stepsize KLMS algorithm. Simulation results for the prediction of chaotic Lorenz series is presented in the terms of mean square error as the figure of merit. To validate the algorithm, comparison is made with KLMS.