Improved Estimation of Embedding Parameters of Nonlinear Time Series by Structural Learning of Neural Network with Fuzzy Regularizer

Abstract

This work proposes an improved refinement scheme of estimation of optimal embedding parameters of a nonlinear time series by a feed-forward neural network trained by structural learning with a fuzzy regularizer (FR). The newly proposed fuzzy rules for tuning regularization parameter enables automatic selection of optimal model with lesser computational load than the basic refinement scheme with RNS proposed by authors earlier. From the simulation results, it has been found that the proposed scheme is very efficient in estimation of optimal embedding parameters in lesser computational time. The short term prediction results also show that the estimated embedding parameters produce better and stable one step prediction.