In this work a new approach to the construction of models and logic circuits of algorithms and procedures for information technology processing and analysis of parametrically related and unrelated data streams within the fractal paradigm is describes. In this case, data streams are defined as information objects whose physical nature can be arbitrary. The information object is investigated apart from any model or scheme, the logical scheme of intellectual technology is built in the form of: facts, regularities and reality. Fractal methods form the framework of the logical, algorithmic and content essence of the approach. The basic premise of the approach is as follows. First, the processing and analysis of parametrically related or unrelated data stream to determine whether it forms a fractal structure and construct a phase portrait of the data stream as an information object. Second, to distinguish the areas of fractal percolation and aggregation in the multifractal structure of the stream, the phase portrait is used. Third, need to estimate the spatial and temporal scales of fractal percolation and aggregation processes.
In the paper a new approach to the construction of models and logic circuits of algorithms and procedures of information technology processing and analysis of parametrically related data streams within the fractal paradigmis is described. The methodology for constructing such models and schemes is based on the construction of the percolation function of a parametrically bound data stream and its information phase portrait. In this case, data streams are defined as information objects whose physical nature can be arbitrary. The information object is not investigated within any model or scheme, but the logical scheme of intellectual technology is built in the form of: facts, regularities and reality. Fractal methods form the framework of the logical, algorithmic and content essence of the approach.
The main premise of the content-semantic essence of fractal methods in information technology processing and analysis of large data streams is as follows. The first stage of data stream processing, is to calculate fractal dimension (geometric and universal) to determine whether it forms a multifractal structure. Second, if the initial data stream is a fractal object, then the stream of integer values of the percolation function is formed. Third, construct and analyze a phase portrait of the data stream; highlight in its structure the areas of fractal percolation and aggregation; estimate the degree of discrepancy between geometric and information fractal dimensions as an indicator of the unity of quantitative and qualitative characteristics of the stream. Fourth, estimate the spatial and temporal scales of fractal percolation and aggregation processes.
The fractal paradigm in the methodology of
development and implementation of information
technologies for processing, analysis and classification
of large data streams, in contrast to traditional methods
and methods [
The paper is structured based on the following assumptions. First, identify and show the fractal properties of information objects. Second, formulate the task of processing and analyzing parametrically coupled data flows based on the fractal paradigm. Third, on the real results show the importance of this work for the development and implementation of IT–technologies for large data flows processing and analysis.
The main and important point in the formulation of data flow fractal properties is the introduction of percolation function concept into the designated subject domain. This concept in this paper is defined as an attribute by means of which data stream fractal properties are denoted and described. It does not in any way correspond to the known probabilistic definition of the percolation function from percolation theory.
To obtain numerical estimates of the geometric
measure of the fractal dimension of a parametrically
related information object as a spatial structure, we used
the well-known Hausdorff–Bezekovich formula [
Let the initial data stream form a metric set М, in which the λ-dimensional outer measure lλ(M) is defined as follows. Considered the ρ-covering of the set M, which is a countable covering of this set with Si sphere of diameter di<ρ, introduce a measure ( , ) = ∑ , (1) finite or infinite, which, as a function of M, is an external measure.
The Hausdorff dimension dim M of the set M is determined by the behavior of lλ(M) not as a function of M, but as a function of λ: = { : ( ) = ∞} = { : ( ) = 0}, (3) that is dimM – is the «transition point»: for λ> dim M, the value lλ (M) = 0, and for λ < dim M, the value lλ (M) –is infinitely large.
Unfortunately, the fractals theory mathematical apparatus based on the fractal dimension of Hausdorff is little applicable to the description of time series and parametrically related information objects. Therefore, to identify patterns due to the properties of the time aspect and the parametric connection of the world events in the information space of the data stream, it is necessary, first of all, to determine the measure of fractal dimension parametric or temporal structure.
For these purposes, the formula for estimating the measure of time structures fractal dimension and Hurst
To estimate parametrically related data stream fractal
dimension measure the following empirical law takes
statistics is used.
place [
= , structures. empirical relation: / =
, where⟨ ( ; )⟩ – is a measure of the time or parametrically related information structure on the interval of time parameter change or the connection Tm, ⟨ (1; )⟩ – for the interval of lengthτ, m – integer number, τ – the duration of the link of the time structure, T – the considered period of time. F –fractal dimension temporary or parametrically associated
On the other hand, Hurst found that the normalized scope R / S for time dependences is well described by the whereR – the scope of the change in the values of the data stream elements over the entire interval T, S –standard deviation, N – the multiplicity factor of the period Tin standard units, H – Hurst index, c – constant.
The Hurst empirical law can be considered as a
special case of the formula (4) for a parametrically
related data stream structure. In this case, the following
analogy is valid: the Hurst exponent can be considered as
an analogue of the fractal dimension estimation F for S =
1. It should be noted that the fractal dimension F and
Hurst index H are not sensitive to such artifacts and
where the lower face is taken over all ρ–covers of the set
M. There is a limit
( ) =
→0
( , ),
(6)
(7)
(8)
phenomena as intermittency in a random medium and
singular errors. This is largely due to the fact that the
above approaches to the data streams fractal structure
analysis do not reflect the information nature of the
objects investigated. Based on these assumptions, we
have obtained a formula for calculating the estimation of
the measure of universal fractal dimension, which is a
synergy of geometric and information dimension [
log ∑ =1 log ∑ =1 1− ij , where pi – the probability of the i–th data stream element ri hitting in the i–th subinterval of △= | length of the subinterval for a given partition of the interval△; ρij – randomized metric between the centers of the j–th and i–th subintervals; rmax and rmin – the maximum and minimum values of the stream elements. – |; ε –
In the work two types of randomized metrics were
considered in the work: geometric and informational. To
calculate the geometric metric of the ρij the following
formula was used [
| | =
– . the ratio below: where | - | – is the geometric distance between the i–th and j–th under the intervals; |r| – is the length of the interval △. To calculate the information metric we used
On the one hand, the above–described formulas for obtaining various estimates of measures of fractal dimensions are integral estimates of fractal properties of the stream of parametrically coupled data. On the other hand, they allow us to formulate and describe the problem of processing, analysis and classification of parametrically linked data stream within the fractal paradigm.
For processing parametrically connected data streams it
is proposed to use mathematical and logical apparatus of
fractal theory [
The primary procedure for processing the original data stream explained by the following logic diagram and
First, the information scale for measuring the values of the elements of the original data stream is determined. In the channels of storage and transmission of information of information–measuring or computing system any data stream is defined as an information object, i.e. binary set. In the language of digital technology, this means that the elements of this set can take two values: one or zero. On the information scale indicated above, either a numerical or information metric can be determined.
Secondly, digitization of the information scale procedure is implemented. The range and price of the scale division are digitized by elements of the natural series.
Third, a stream of integer values of the percolation function of the original data stream is formed. The percolation function reflects and describes the geometric structure of the "leaky information space" of the original data stream.
Fourth, the procedure of constructing a phase portrait of the data stream is implemented. The phase space is a plane on which the following coordinate system is determined, namely: the abscissa axis - percolation function values, the ordinate axis – digitized information scale values.
The relationship between the scale integer values and another scale of measurement of the source data elements stream is carried out by following attributes: • general scale of variability of the values of the elements of the original array in any other noninteger algebraic system of their measurement; • common scale division price in a non-integer algebraic system; • number of significant digits.
The integer values of percolation function are determined by following ratio:
ℎ = − −1 , where hi – values of percolation function, ri – the number of the interval in which the corresponding element of the original data stream falls.
The number of partitions L of interval △= | – | into subintervals is determined based on the following ratio: (9) (10) =△/ .
The main premise of the problem statement is to determine whether the processed and analyzed data stream is a fractal? If so, describe its fractal structure and calculate the integral estimates of fractal dimensionmeasures. The solution of problem is to obtain above estimates of fractal dimension measures and construct the phase portrait.
Two data sets for processing and analysis using fractal methodswere used. In the first case, data were presented by the stream of regular harmonic function values. In the second case, were used empirical data presented by the icm–20608 quadcopter gyroscope measurements.
To solve this problem, a software component that implements the above algorithm of data stream processing was developed. The function y = 2.5sin(2πt) was used as a real harmonic function that induces the data stream. The volume of data stream was not less than ten thousand elements.
Total scale for the harmonic function varied within (– 2.5;2.5).Two different scale division intervals ε=0.01 and ε=0.001 were taken. The number of partitions L=500 and L=5000 respectively. Gyroscope data stream values were changed in the range (–840;1270). Scale division intervals ε=0.01 and ε=0.001.The number of partitions L=211000 and L=2110000 respectively.
As a calculations result, the following fractal dimension measure estimates were obtained, which was determined by the formula (6) for various metrics ρij, which were determined by the formulas (7) and (8).
For the harmonic function (figure 1) the following numerical estimates of universal fractal dimension for various metrics ρij were obtained: 1. Geometric metric: db = 0,950 for ε =0,01; db = 0,827 for ε =0,001; 2. Information metric: db = 1,060 for ε =0,01; db = 0,837 for ε =0,001.
The above measure of universal fractal dimension values db show that the harmonic function reflects the information set, which has a regular structure (the condition of regularity is db→1). Here the geometry of information set and its information compendency are in good agreement. This is well confirmed by the fact that the db values for geometric and information metrics are close for different ε values.
Figure 1 Graph of 2,5sin(2πt) function
Another information picture is observed for the gyroscope values, which are shown in figure 2. The numerical estimates of the universal fractal dimension for the set of gyroscope data using a geometric metric are given below: db = 3.438 for ε =0,01; db = 7.439 for ε =0,001.
The db modulus value with decreasing ε increases significantly, which is the criterion and indicator of irregularity. The designated information object is a data stream with an irregular structure.
The values of percolation function 2,5sin(2πt) for different values of common scale and tick marks are shown in figures 3 and 4. As can be seen from figures 3 and 4 at lower values ε percolation function exhibits pronounced properties of regularity or continuity for 2,5sin(2πt) function calculated values variability. A geometric illustration of percolation function for gyroscope data stream is shown in figure 5. The graph of percolation function fully reflects the properties of irregularity and the data stream elements values singularity.
This stage of parametrically related stream data processing and analysis in the information technology logical chain allows us to obtain data stream fractal nature and numerical estimates of fractal geometry measures, as well as to obtain percolation function values stream. The second stage in the logical chain of information technologies for processing and analysis of parametric bound data stream is the construction of its phase portrait.
Let's illustrate the results obtained at this stage, using the example of data streams considered earlier.
For the data stream induced by the harmonic function 2,5sin(2πt) for ε=0.01 and ε=0.001, phase portraits are shown in figures 6 and 7.
Phase portrait shown in figure 6 illustrates following fractal nature properties of analyzed data stream. First, at a given General scale and the scale division intervals, the distinct fractal properties of stream do not appear, but with an increase in the values of ε, it will already have these properties. Secondly, the presented data stream has the property of regularity, because the framework of the phase portrait has a closed trajectory. Figure 7 shows a phase portrait of the same data stream for a smaller value of ε. The geometric image of this portrait illustrates quite fully regular properties of data stream in question. The framework of the geometric image of the phase portrait forms a pronounced closed trajectory.
The phase portrait shown in figure 8 is quite clearly and fully illustrates stream multifractal structure in phase space which presents fractal percolation and aggregation region. The space of percolation processes dominates in the region of small values of gyroscope readings (in the phase portrait – a "clot" of points in the center).
The geometry and topology of this phase portrait region reflects a regular structure on the icm–20608 quadcopter gyroscope values set. Such structure is typical for a stable and regular operating mode. Fractal percolation covers the area of large gyroscope readings.
On the one hand, this region of the phase space reflects the singular processes in the icm–20608 gyroscope. The dynamics of a quadcopter in this case occurs along a complex trajectory, which fits into the regular mode of its operation. In Figure 8, this feature is reflected in the form of sharp changes in the values of the gyroscope readings and is indicated by arrows.
On the other hand, it allows to reflect and to describe the spatial and temporal scales of the non-standard or non-stationary gyroscope operation mode. The phase portrait shown in Fig. 8 reflects gyroscope stationary operation mode with singular phase transitions and no disturbances. Full-scale experiments on modeling the non-stationary operating mode of the gyroscope were not carried out due to objective reasons for the authors. Laboratory studies of this mode of operation and analysis of the results allow us to draw preliminary conclusions. In this case, the intermittence of the processes of fractal aggregation on fractal percolation trajectories with a positive and negative gradient will be observed.
As we can see spatial and temporal scales of the fractal percolation and aggregation processes are sufficiently fully and substantially illustrated by the phase portrait.
The results of processing and analysis of parametrically related data streams, indicated above, allow us to make a number of conclusions and generalizations.
First, fractal methods in large parametrically linked data streams processing and analysis is based on logical schemes of their phase portraits cognitive analysis, decoding of information hidden in them are promising and unique paradigm in the information technologies and smart information–measuring systems development.
Second, the streams of parametrically related data can be processed using various processes and methods of fractal theory and genetic data for both the collection and population of sample data and their analysis. These methods and processes reflect and define the features of the resulting estimates of fractal measures and dimensions, as well as the scope of the conclusions that can be drawn from these data. In this case, stream phase portrait are parametrically associated to the data stream. Phase portrait of the data stream determines and describes the regular and irregular properties of its structure relative to the information scale of measurements.
In a wide aspect of fundamental research in the field of intelligent information–measuring systems and intelligent information technologies, the results of this work for the first time allowed us to show how and in what the synergy of such entities as facts, laws and reality is manifested. Is it possible to draw such analogies in the framework of traditional models, algorithms, schemes, etc.? If yes, then show the results of the identified analogies and formulate trends of their theoretical development and practical continuation.
Applied aspects of the results are closely related to the solution of a wide range of problems in the field of physical experiment, development and implementation of information technologies for control, diagnosis and control of nuclear power plants, and many others. On the one hand, the methods of fractal theory of solving complex nonlinear problems of processing, analysis and interpretation of the results of physical, biological and medical experiments are proposed. On the other hand, a new IT – technologies was developed and implemented in the trend of DAMDID processing, analysis and classification of parametrically related data.
For software implementation of IT – technologies were used data from the icm–20608 quadcopter gyroscope.