In order to develop the scientific and technical foundations of creating an artificial intelligence system for monitoring tectonic origin emergencies, the paper in the results of modeling and forecasting conducted on the basis of neural network technologies, total number of occurrences, with discretion in one month, earthquakes on Earth as one of the main characteristic indicators of the seismic activity of the Earth's globe are presented.
The tendency to an abrupt increase in the number and destructive power of natural
disasters over the past few decades of the life of society leads to a deterioration of
socio-economic and environmental consequences. It indicates the need to develop
effective measures to prevent and eliminate emergencies (ES) of various nature on the
Globe [
A promising direction for solving this problem is the development of an effective
hazard detection system at the stage of their inception. Also, the causes will be
establishing of the occurrence these factors manifestations and effects on them in order to
prevent the occurrence of emergencies. This has been implemented on the basis of the
classical control loop presented in Fig. 1 [
This article is part of a planned set of scientific studies aimed at developing a
safety system. This eliminates or minimizes losses as much as possible under
conditions of manifestation of an emergency. The work is focused on studying the
processes of emergence and spreading of emergencies of lithospheric origin, which
represent or may pose a serious danger to the life of society [
Determination of
parameter Modeling and prediction of development of emergency Analysis and systematization of information Data processing
The person making the decision
CHOICE Management decisions for the prevention and elimination of emergencies
SYSTEM PERFORMANCE
SOLUTIONS Formalization of management decisions Bringing a decision to the performers
Hazard registration
ES LOCAL TERRITORY
Influence on the hazard
When solving the problem of creating an artificial intelligence system for monitoring emergencies of the tectonic nature, there is a need to studying and simulation the processes of occurrence and propagation of the seismic activity level of the local territory under the conditions of seismic activity of the Earth’s globe, as an element of the system of nonlinear energy interactions Sun – Earth – Moon. 2
Development of scientific and technical bases for creation of artificial intelligence system for monitoring of tectonic origin of the emergency, which is realized by constructing neural network models of forecasting the level of seismic danger of the Earth’s territory by the amount and destructive energy of the emergency of tectonic origin.
The construction of neural network models for predicting the level of seismic
danger in the Earth’s globe was performed using the STATISTICA statistical package
[
Today, the basis of predictive observations of Earth’s seismic activity is knowledge
of the physical laws of the earthquake mechanism and control of the physical fields in
the seismic zone [
The mechanistic approach, which has been developing in various countries for
quite some time, has provided answers to many questions about the preparation of the
earthquake. Currently, more than a dozen earthquake models have been developed,
the most famous of which is following papers [
a model of avalanche-unstable crack formation (AUC), which consists in the rapid growth of the number of cracks, their interaction with each other and eventually the occurrence of a main or main rupture, the shift of which instantly drops the accumulated elastic with the formation elastic waves;
dilatant diffusion model (DD), in which crack formation also plays a decisive role, but unlike the AUC model, fractures occur and the presence of water in the rocks of the epicentral region is essential;
consolidation model of earthquake preparation (Dobrovolskiy’s model) describes earthquake preparation as a process of emergence and evolution of rigid inhomogeneities in a continuous environment, as well as some other models that emerged as a generalization of the study of stress and fracture processes of solids samples in the laboratory and their subsequent transfer to natural seismic events.
At the same time, the following methods are now used to obtain a comparative
assessment of the level of life-threat in the conditions of manifestation of a emergency
[
statistical, based on the analysis of emergency statistics on local territories over several years to determine the risk indicators;
probable, based on the application of mathematical models, which connects the prerequisites for the appearance of emergency with the possibility of their manifestation; expert based on expert judgment in combination with fuzzy set theory.
The advantage of the statistical method is objectivity. Probable and expert methods allow to take into account the sources of potential danger, which are rare in the form of emergencies, but the consequences of which are catastrophic. However, the probable method is extremely cumbersome and time-consuming, requiring a large number of outputs, which results in low accuracy of the results obtained. In the absence of tried and tested mathematical models and reliable enough initial data for them, it is advisable to carry out an expert evaluation of the impact on the possibility of implementing large-scale emergencies of a large number of difficultly formalized initial data.
The prospects of using a statistically probabilistic approach to predict earthquakes
in a separate Earth globe, without taking into account the external factors (as an
element of the Sun – Earth – Moon nonlinear energy interaction system), are given
in [
Thus, the use of neural network technologies is one of the promising directions for the
development of approaches to predicting earthquakes across the globe into systems of
nonlinear Sun – Earth – Moon interactions. This determines the direction of our scientific
research in the field of monitoring of the tectonic origin emergencies [
The dynamics of the physical processes of the Sun – Earth – Moon system affecting the seismic hazard level of the local territory functioning can be schematically represented according to Fig. 2 – 4. This can be characterized by the following spatial constructions within the solar galaxy.
1. The axis of rotation of the Earth in the celestial sphere describes a complex
wave-like trajectory. The points of the axis of rotation are at an angular distance of
about from the pole ecliptic (Fig. 2). The vertex of the cone coincides with the Earth
center. The points of equinoxes and solstices move along the ecliptic towards the sun.
Moments of gravitational forces influence on the equatorial bulges and vary
depending on the positions of the Moon and the Sun relative to the Earth. When the Moon
and the Sun are in the plane of the Earth’s equator the moments of forces disappear. If
tilts of Moon and Sun are the maximum, then the magnitude of the torque will be
greatest. The nutations, owing to fluctuations in the moments of the forces of the axis
of rotation of the Earth have been observed by consist of a series of small periodic
oscillations. The main nutations have a period of 18.6 years - the time of the orbital
nodes of the Moon. Movement with this period occurs on an ellipse. The major axis
of the ellipse is perpendicular to the direction of the precessional motion and is equal
to; small - parallel to it and equal. Next in magnitude of the amplitude are the
components with a period of 0.5 year, 13.7 days., 9.3 years, 1 year, 27.6 days. etc., therefore
the trajectory has the form of “thin laces” (shown on the enlarged fragment in the left
part of Fig. 2) [
2. The pressure from the solid inner core and the surrounding melt (outer core)
onto the mantle arises as a result of the eccentric revolution of the Earth’s shell
around the displaced inner core, which squeezes the shell from the inside. The forces
compressing the shell of the sphere (planet) and drawing it inward to the core arise in
other parts of the planet. This process has two components: impact at the expense of
the annual displacement of the center nucleus relative to the center of the globe
(Fig. 2 – 4); impact at the expense of eccentric circulation of the core relative to the
lower mantle, when due to the difference in angular rotation velocity of the core and
lower mantle ( 1 – angular velocity of rotation of the mantle; 2 – angular velocity
of rotation of the outer core; 3 – angular velocity of rotation of the inner core;
2 1 – angular velocity of rotation of the outer core relative to mantle
(“western drift”)), therefore, there are zones of high pressure and vacuum
( P1 P2 where P1 and P2 are indicators of pressure of the inner core of the globe on
its surface), affecting the level of seismic activity of the surface of the Earth (Fig. 3).
As long as there is a difference in the angular velocity of rotation and displacement of
the nucleus, the appearance of such zones will be maintained [
P1 1
3. Internal elastic stresses arise in the process of moving lithospheric plates (Fig. 4), which are energy sources of earthquakes. The occurrence depth of elastic stresses depends on the nature of a movement plates. The relative motion of lithospheric plates leads to the emergence of shallow (not deeper than 20–25 km) earthquake sources and dipping of lithospheric plates into the mantle initiates the appearance of sources of deep (exceeding 70 km) earthquakes. The probability of elastic stresses - sources of earthquakes decreases with increasing distance from the interface of the lithospheric separation plates.
4. Surface and bulk seismic waves are the propagation factors of earthquake
hazards Z0 , that can cause secondary earthquakes [
5. The probability of mutual amplification or weakening of bulk seismic waves
increases in the process of spatial-vibrational movement of the Earth’s internal core and
its effect on the external core. Consequently, the possibility of secondary earthquakes
Z increases also [
7. The territorial-temporal changes in the intensity of the natural electromagnetic
field pulses of the Earth initiating anomalous processes in the atmosphere occur due
to the movement of the Earth’s inner core has been established [
Directions of movement lithospheric plates (arrows) Elastic stresses
Plate І Lithospheric plate section
Z’ Bulk seismic waves
Mantle 1 2
Earthquake probability
Z0 3 Outer core
Inner core
Earthquake
Surface seismic
waves Z’’ Plate ІІ x
Thus, combining the analysis results of the impact dynamics and energy of the internal physicochemical processes of the Earth on the origin generating tectonic processes allowed to formulate an approach to studying the nature of seismic phenomen. It is an important tool for analyzing the results of civil defense research on the development of models for the development of ES tectonic nature.
According to the purpose of the research, the solution of the scientific problem in the work is ensured by constructing an artificial neural network (ANN) model – a model for predicting the level of seismic activity of the planet from the conditions of the system functioning of nonlinear energy interactions Sun – Earth – Moon. This is a time series model, where the initial indicator, according to Fig. 5, is the total number of earthquakes in the world N t (provided that M 5 the magnitude of the earthquake), depending on the current time of analysis ( t ), as well as the distance of the Earth’s inner core from the center of the planet t and the change in the length of day
LODt St 86400s (where
St r0 86400s – the length rt of day, r0 7,292115 105 rad s – constant (average) angular velocity of the Earth’s own rotation).
For the development presented in Fig. 5 models have a multilayer perception
(MLP) selected. The input parameters of this model are the results of the analysis of
the monthly dynamics of indicators, N t, t and LODt for 2009 – 2018,
which are presented in [
February
May
When developing a mathematical model for predicting the level of seismic activity of the Earth ball, based on the decision parameter N t on the complexity of the MLP architecture, it was based on five analyzes of the learning results of networks, which accidentally included five hundred neural networks. The criterion for choosing the optimal network was the relationship between the forecast error and the complexity of the architecture. The results of the analysis are presented in Table 1.
Study productivity 0,944 0,935 0,939 0,938 0,967
Control productivity 0,859 0,634 0,859 0,884 0,727
For neural network training, all observations were divided into three samples. By default, random sampling was performed between samples to avoid retraining the network and to ensure quality generalization (forecasting). The first sample (Educational – 50% of observations) was used to train the network; the second (Control – 25% of observations) – for cross-validation of the learning algorithm during its operation; third (Test – 25% of observations) – for the final independent testing of a trained neural network. The training was done with speed 0,01 .
Presented in Table. 1 networks are characterized by a more effective balance between modeling error and architecture complexity for time series analysis of the forecast of the monthly dynamics of Earth’s seismic activity ( N t ). This was the basis for further construction of the three-layer MLP 15-12-1 network, which has fifteen inputs (Fig. 5), 12 elements in the hidden layer and one logical output function of thirteen inputs, presented in Fig. 7.
Fig. 7. MLP 15-12-1 Three-Layer Perceptron Architecture with Logical Signal Transmission for Time Series Prediction of Globe Seismic Monthly Dynamics ( N t ) (Source: author's own elaboration)
In this case, the use of logical activation functions, with scaling parameters, was based on a given fraction of span of a logical function equal 0,9 (respectively range 0,05; 0,95 ) to the range of training of the neural network. The function of activating the hidden layer of the perceptron MLP 13-10-1 is hyperbolic tangent. This allows for a slight extrapolation of the data. In addition, the use of logical functions stabilizes learning.
The results of checking the adequate prognostic performance of MLP 15-12-1 network are presented in Fig. 8, where the dependence observed ( N* t ) statistic the values of the Earth’s seismic activity indicator via predicted ( N t ) values. The coefficient of correlation between these indicators on the results of the training "Educational Choice" is equal rN2* tN t 0,967 . values of the Earth’s seismic activity indicator via predicted ( N t ) values by the MLP 15-12-1 network is presented in Fig. 9.
According to the data analysis Table 2 it is necessary to state that obtained within the limits of the ideas about the dynamics of physical processes that occur in the system Sun – Earth – Moon and affect the level of seismic danger of functioning of the local territory of the planet (see Fig. 2 – 4), the MLP neural network 15-12-1 and the results of its forecast allow us to ascertain the adequacy, in accordance with the data of Figs. 8 and the rN2* tN t 0,967 parameter figure presented in Fig. 5 models for predicting the level of seismic activity of the globe. LOD*t 0,81 1,00 0,96 0,98 0,70 0,19 0,01 0,05 0,17 0,27 0,37 0,24 06 level rN2* tN t 0,967 .
The results obtained are the basis for the development of scientific and technical bases for the creation of a system of artificial intelligence for the performance of tasks of monitoring of tectonic origin. 45. Garcia, R., Crespon, F., Ducic, V., Lognonne, P.: Three-dimensional ionospheric tomography of post-seismic perturbations produced by the Denali earthquake from GPS data. Geophys. J. Int., 163, 1049 – 1064 (2005). 46. Heki, K., Ping, J.: Directivity and apparent velocity of the coseismic traveling ionospheric disturbances observed with a dense GPS array. Earth Planet. Sci. Lett., 236, 845 – 855 (2005).