• A
  • A
  • A
  • ABC
  • ABC
  • ABC
  • А
  • А
  • А
  • А
  • А
Regular version of the site
Menu
  • HSE University
  • Publications of HSE
  • Book chapters
  • Выделение периодических компонент в многомерных точечных процессах при воздействии нескольких регулярных факторов

Book chapter

Выделение периодических компонент в многомерных точечных процессах при воздействии нескольких регулярных факторов

С. 95-100.

Investigations of dynamic properties of complex nonlinear systems are a very promising area of modern science. Information about behavior of such system is often fixed as multidimensional point process where every event is a point in a space-time volume. If a time of observation is limited and a number of events in the dataset is no more than several tens the process can be perceived by an external observer as a random even if this process is due to the actions of a few regular factors or the interactions between of the elements of the complex system. A very small number of events, adding a certain amount of truly random events, fluctuations in time and location of the events are leads to the ineffectiveness of traditional methods of processing noisy signals.

This paper presents a method for the analysis of multivariate point processes based on a given physical model of the interaction of the system elements and \ or the joint effect of several external factors that have a regular character. One-dimensional model of a point process was analyzed in detail, the possibility of generalization of the proposed method to the multidimensional case was showed. The main idea of the method is to identify in the event flow quasi-periodic chains of events characterized by a high degree of non-randomness. A periodic slit mask (similar to Dirac comb) was used to find the subset of events. Of all found chains were chosen only chains with  the non-randomness level of above the some threshold.

Artificially generated point process was used to compare of the proposed method with the method based on an assessment of the harmonic modulation degree of the Poisson point process intensity. Test event occurs if at least three of the six harmonic oscillations with different phases and periods overlap near to zero level. This dataset was consisted of 79% and 21% deterministic and random events respectively. It was found that the traditional method failed to reveal five of the six periodic components that have been used for generating test point process. In contrast, the proposed method based on estimating the degree of non-randomness of quasi periodic chains reliably detects all the original periodic components. More over, it was shown that the proposed method provides a much better sensitivity and resolution. Errors in determining of test harmonic oscillation periods were less than 0.057%. Note that along with real components in the spectrum of the point process is fixed many other peaks are associated with the fractional multiples of the periods of quasi-periodic chains. These peaks are similar to harmonics in the standard spectral analysis. Availability of information about events that are part of quasi-periodic chains and high resolution presented method can accurately detect and identify such fractional multiple harmonics.