The Prediction of Non-Stationary Physical Time Series Using the Application of Regularization Technique in Self-organised Multilayer Perceptrons Inspired by the Immune Algorithm

Neural networks have been widely used in nonlinear time series prediction. They have generated lot of interest due to their comprehensive adaptive and learning abilities. Neural networks have been used in Medical forecasting, Exchange rate forecasting, stock index prediction, and other areas, which show a practical value of neural networks. This paper presents a novel application of the Self-organised Multilayer perceptrons network that is inspired by the Immune Algorithm (SMIA) in physical time series prediction. The Regularization technique is used with the self-organised multilayer perceptrons network that is inspired by the immune algorithm (R-SMIA). The results of 20 simulations generated from two non-stationary physical time series using various neural networks are demonstrates. The results of R-SMIA were compared with four networks which include the MLP, R-MLP, FLNN, and SMIA networks.