Robust identification and prediction using Wilcoxon norm and particle swarm optimization

The paper introduces a novel method of robust identification of complex plants and prediction of bench mark time series. It is assumed that training samples used contain strong outliers and the cost function chosen in the proposed model is a robust norm called Wilcoxon norm. The weights of the models are updated using population based PSO technique which progressively reduces the robust norm. To demonstrate the robust performance of the proposed technique standard identification and prediction problems are simulated and the results are compared with those obtained by conventional MSE norm based minimization method. A significant improvement in performance is observed in all cases.