Initialization of multilayer forecasting artifical neural networks

In this paper, a new method was developed for initialising artificial neural networks predicting dynamics of time series. Initial weighting coefficients were determined for neurons analogously to the case of a linear prediction filter. Moreover, to improve the accuracy of the initialization method for a multilayer neural network, some variants of decomposition of the transformation matrix corresponding to the linear prediction filter were suggested. The efficiency of the proposed neural network prediction method by forecasting solutions of the Lorentz chaotic system is shown in this paper.

Training Algorithm for Neuro-Fuzzy Network Based on Singular Spectrum Analysis

In this article, we propose a combination of an noise-reduction algorithm based on Singular Spectrum Analysis (SSA) and a standard feedforward neural prediction model. Basically, the proposed algorithm consists of two different steps: data preprocessing based on the SSA filtering method and step-by-step training procedure in which we use a simple feedforward multilayer neural network with backpropagation learning. The proposed noise-reduction procedure successfully removes most of the noise. That increases long-term predictability of the processed dataset comparison with the raw dataset. The method was applied to predict the International sunspot number RZ time series. The results show that our combined technique has better performances than those offered by the same network directly applied to raw dataset.