Fractal analysis of nonlinear systems and the construction of its basis of predictive neural networks

The algorithm pre-fractal analysis of chaotic time series for the construction of predictive neural networks (PNN), based on the method developed to identify the time lag. The algorithm was applied to the temporal samples from different areas of science and technology: economics, medicine and electronics. The method is based on fractal analysis of one-dimensional output signal coming from the generating system provided in the form of black box. Sequence analysis of the following: definition of the time lag of a number and its phase shift forming a time series, the restoration by the method of delays attractor in the space and the calculation using the method of Grassberger - Procaccia its maximum embedding dimension and the correlation fractal dimension Made in the reconstruction of the attractor space by the method of Takens. Next crating PNN: selects its structure and principle of implementation. Education networks held by the gradient method Levenberg - Marquardt. The results of predictions obtained from a network built on the basis of multilayer perceptron with one hidden layer.

fractal analysis, time series, the lag time, predictive neural network training Levenberg - Marquardt method