The processes of intellectual monitoring in emergencies are studied. The intelligent monitoring system is an environment for creating and using intelligent agents to provide knowledge of decision-making processes. In emergencies, objects acquire new properties quickly, and the informativeness of the results of previous observations decreases. To increase the power of data mining tools, monitoring agents are combined into agent functionalities with a multi-tier structure. The paper presents the results of research on the processes of formation of multiechelon polyagent functionals. The efficiency of construction of a multi-echelon polyagent functional in solving the problem of predicting the incidence of the population of Ukraine on Covid-19 in conditions of low informativeness of the results of observations has been experimentally confirmed.
In the context of crisis monitoring, reducing the informativeness of the observations results is one of the problems that reduces the efficiency of the process of extracting knowledge from data sets. In emergencies, the monitored objects move to another state and acquire new properties. These properties are not fully reflected in the arrays of the results of previous observations. The period of time for new observations is much longer than the time during which it is necessary to provide the results of monitoring to the decision maker.
This problem is overcome by information
technology of intelligent monitoring through the
use of a multi-agent approach to the creation of
monitoring information systems (MIS). A
separate agent is built to perform monitoring
tasks. When the informativeness of the input data
arrays (IDA) is reduced, superagent formations
are built - agents functionals. The concept of
"functional" is interpreted here as "function of
functions". Agent functionality (AF) is a
structural element of the monitoring information
system [
The technology of building polyagent
functionalities (PAF) involves the creation of
agents with structural tasks and combining them
into a system based on a matching IDA. Matching
IDAs for structural agents are formed on the basis
of the same array of observation results [
This paper presents the results of research on the process of forming a many echelon structure of the polyagent functional of the monitoring information system. The technology of construction of PAF is presented on an example of improvement of process of performance of the monitoring task on forecasting of number of diseases of the population of Ukraine on Covid19.
There are several approaches to creating
many-agent systems. The problem-oriented
approach is based on the assertion that several
agents can achieve a goal that is beyond the power
of one agent [
The object-oriented approach assumes that
MAS is a combination of autonomous intelligent
agents, each of which performs its task and each
interacts with other agents of this MAS [
The problem-oriented approach to the creation of MAS is used by information technology of intelligent monitoring (ITIM) in the process of its implementation in the form of MIS to perform the task of predicting the incidence of the population of Ukraine on Covid-19.
In case it is not possible to build an agent to
perform a new monitoring task due to insufficient
informativeness of the array of results of previous
observations, the agent functional is built. A
system approach is used to build the agent
functionality. Agent functionality is built as a
system. The functional emergence is formed due
to the effective combination of agents and a more
complete reflection of the properties of the object
in the model knowledge bases of the monitoring
information system. This effect is manifested by
improved signal characteristics at the MIS output
and increased adequacy of interpretation of
monitoring results. [
An attempt to build an agent with a monitoring
task to predict the number of diseases in Ukraine
with a horizon of 7 days with this forecast task
was unsuccessful. The array of observation results
obtained from an open source [
MIS did not have built agents that would perform tasks based on different information sources in other subject areas, so the construction of multi-agent functionality was impossible. To fulfill the monitoring task, the IIA built the PAF.
The construction of a single-echelon PAF
allowed to obtain the forecasting result with an
error of 13.99% on the horizon of 7 days and to
improve the average forecasting error to 6.15%
[
statement description and tasks
To reliably assess the influence of factors, the
signs of which are included in the array of
observations, it is necessary to build a model with
a minimum value of forecasting error. Therefore,
studies of the process of constructing a PAF to
enhance the emergence of an agent combination
were performed. The task of building a PAF is to
create a method that would provide knowledge
about the patterns of pandemic development in
Ukraine in the future from previously observed
results. The forecasting problem formulated in [
An array of X results of population morbidity monitoring during 2020 is given:
X = {xij}, i=1,n; j=1,m , (1) where n is the number of signs that reflect the incidence of the population, m is the number of observation points (recorded number of diseases in countries with a discreteness of 1 day).
The number of observation points is determined by the duration of the historical period of time during which the values of morbidity were recorded:
T = {t, t-1, t-2, t-m}, (2) where t is the observation time; m is the number of observation points.
The monitoring information system builds a set of agents with structural tasks:
Y = {Yx1, Yx2, Yx3, …, Yxn}, (3) where n is the number of agents that perform structural tasks.
It is necessary to build a polyagent functional for predicting the incidence of Covid-19 population of Ukraine with a forecasting horizon of 7 steps, for which the deviation of the forecasting results from the actual values becomes minimal:
Zx1= f(T, X, Y, t+7), (4) where Zx1 – signal at the output of the agent functional with the monitoring task of forecasting the number of diseases of the population of Ukraine; t+7 – forecast horizon (7 steps).
output signals of the agents included in the structure of the lower echelons of the functional.
Execution of new monitoring tasks on the basis
of previous results of supervision is provided by
construction of agent functionalities with
multilevel hierarchical structure. The upper
echelons of the functional are formed from agents
that are not included in the structure of the lower
echelons. Increasing the informativeness of the
signals at the output of these agents is achieved by
increasing the number of features in the agent IDA
due to the signals from the output of lower
echelon agents. The hierarchical structure of
multi-echelon agent functionals is built by the
method of ascending synthesis of elements [
The IDA for the structural tasks of the higher
echelon agents is formed from the features
obtained as a result of observations and from the
Table 1
Signs of the initial description of diseases on COVID-19 [
Indicators Observation time; morbidity in Ukraine; morbidity in France; morbidity in Belarus; morbidity in Georgia; morbidity in Germany; morbidity in Israel; morbidity in Italy; morbidity in Moldova; morbidity in Slovakia; morbidity in Slovenia; morbidity in Russia; morbidity in Portugal; morbidity in Poland; morbidity in Romania; morbidity in Spain; morbidity in Turkey; morbidity in Egypt; morbidity in Greece; morbidity in the United States; morbidity in China; incidence in England
To perform the monitoring task of forecasting
the number of diseases of the population of
Ukraine on Covid-19 on the horizon of 7 days, an
array of signs was formed, obtained as a result of
observations of morbidity of the population of
other countries during 2020 with a step of 1 day.
The results of observations were obtained from an
open source [
Comments
results of The daily observations obtained during 2020
After that, the method of forming the structure of a multi-echelon polyagent functional ascending construction of layers was applied. Agents who did not complete their task on the lower layer of the PAF began to perform this task on the upper layer. All agents that performed the task formed the structure of the layer. The signals at the output of these agents were used as additional features in the input array of higher echelon.
Table 2 presents the characteristics of the output signals of agents that performed structural tasks in the construction of PAF layers. The characteristics of the agents that performed the structural task and entered the structure of the corresponding echelon in table 2 are highlighted in bold.
Thus, after the construction of the polyagent functional, the error in predicting the incidence of the population of Ukraine decreased by 72.96% compared to the results of forecasting this indicator by the agent who performed this task.
The input array for the upper echelon was formed from the signals at the output of the agents that entered the structure of the lower echelons.
If the error in predicting the signal at the output of the agent is less than the limit value of 12%, the structural task is considered completed, the agent acquires the state "Used" and is included in the structure of the corresponding echelon. Structural tasks that were not performed by the lower echelon PAF agents are assigned by the MIS to be performed by the upper echelon. According to Table 2, 9 structural tasks for morbidity forecasting in the respective countries were performed at the first echelon. The structure of the first echelon of PAF includes 9 agents. For execution on the second echelon MIS transferred 7 structural tasks. Of these, 3 tasks were completed.
And, accordingly, the structure of the second echelon was formed by 3 agents. The prediction errors in the signals at the output of all agents were less than the characteristics of the signals that had agents with the same tasks in the previous echelon. Therefore, unfulfilled tasks are transferred for execution to the highest echelon.
Of the 5 tasks of the third echelon, the agents did not complete any. Agents 4 and 19 at the output had prediction errors greater than those they had in the previous echelon.
Prediction of morbidity in Ukraine on the 7th day Prediction of morbidity in France on the 7th day Prediction of morbidity in Belarus on the 7th day Prediction of morbidity in Georgia on the 7th day Prediction of morbidity in Germany on the 7th day Prediction of morbidity in Israel on the 7th day Prediction of morbidity in Italy on the 7th day Prediction of morbidity in Moldova on the 7th day Prediction of morbidity in Slovakia on the 7th day Prediction of morbidity in Slovenia on the 7th day Prediction of morbidity in Russia on the 7th day Prediction of morbidity in Poland on the 7th day Prediction of morbidity in Portugal on the 7th day Prediction of morbidity in Romania on the 7th day Prediction of morbidity in Spain on the 7th day Prediction of morbidity in Turkey on the 7th day Prediction of morbidity in Egypt on the 7th day Prediction of morbidity in Greece on the 7th day Prediction of morbidity in the United States on the 7th day Prediction of morbidity in China on the 7th day Prediction of morbidity in England on the 7th day Prediction of morbidity in Canada on the 7th day 1 11,01% 4,43% 13,81% 5,50% 69,94% 3,56% 7,66% 101,38% 197,22% 14,18% 1,94% 26,86% 2,94% 8,80% 4,70% 91,68% 48,90%
In the fourth echelon, structural monitoring agents managed to complete the task of predicting morbidity in Slovenia. The construction of the fifth echelon of the PAF has made it possible to carry out morbidity forecasting tasks in England and Canada.
The output signal of the 6-tier polyagent functional performed the task of predicting the number of diseases in the population of Ukraine on the 7th day after the last observation. The forecasting error on the horizon of 7 steps became 4.57% with the average value of 7 steps - 2.6%.
The construction of multilevel polyagent functionalities allows overcoming the problem of performing monitoring tasks in the conditions of insufficient informativeness of the arrays observation results. Using the results of previous research on the construction of echelons the agent functional and feedback in their structure, a method for forming a multi-echelon structure of polyagent functional is proposed.
The experimental test of the method was carried out in the process of solving the task of forecasting in the conditions of insufficient informativeness of historical data.
The results of the solving problem are predicting the incidence of the population of Ukraine on Covid-19 in conditions of low informativeness. Due to the construction of a multi-tier polyagent functional, the prediction error is reduced by 72.96%.