time series

The article addresses the problem of the restoration of missed values in the results of testing the recipients given by the time series. As experimental data, the time series with spaces are given. Recovery efficiency is estimated by the relative error of the recovered value. Examples of restoration missing data in the time series table and the individual time series are given. Used simple methods for replacing missed value by average, weighted average and median.

The recurrence plot concept was disclosed as a proximity matrix between phase space depicting points on phase area for time series. Recurrence plots construction equations and application results of suggested method are given in the MS Excel. Three types of signals: harmonic, meanders, triangular are used as a graphical material and even noise distribution is imposed on this signals. Results have significant practical value, because recurrence plots method is not widely spread in Ukraine, however it is easy to implement in spreadsheet.

Analysis of the debris flow formation factors which cause the long-term activity of debris flows is made. The methodology of the debris flows prediction subject to meteorological, hydrological, seismic, heliophysical factors is proposed. The regularities of long-term seasonality of these factors by using autocorrelation and spectral analysis are revealed. The integral rate of probability of debris flow intensification is calculated. The time series of this integral rate is extrapolated and the following peak of debris flows activation is predicted.

Presented a model of carrier personnel from experimental results. Model of the filed two components – a trend equation and the distribution of deviations from trend data. Trend equation is analytic function – with mixed polynomial degree. Rejection levels approximated by Rayleigh distribution law. Both models provide a system for objective assessments identifying personnel.

In this paper the radial-basis neural network was used for software failures prediction. The influence of activation function of the RBF neural net on the learning efficiency and software failures prediction is studied. It is shown that the optimal activation function is Inverse Multiquadric with 10 neurons in the input layer and 30 neurons in the hidden one (square of Pearson correlation coefficient is 0.997 and mean deviation is 14.4).

The recurrence plots tool origination and foundation process is considered. Construction method of recurrence plots based on time series is described. Recurrent plots dependence on the input parameters is analyzed and shown.

The article suggests and describes a GMDH algorithm for the synthesis of co-operative immune network in the solution of tasks of forecasting of time series. Conducted comparative experiments have shown that the use of external criteria improves adaptability, robustness and accuracy of the obtained solutions.