Model for Monitoring of Socio-Economic Processes Using Fuzzy

Cognitive Map and Algorithms for Detecting Structural Changes

Zinaida K. Avdeeva

, Elena A. Grebenuyk

and Svetlana V. Kovriga

V.A. Trapeznikov Institute of Control Sciences of RAS, 65 Profsoyuznaya street, Moscow, Russia

Keywords: Monitoring, Structural Shifts, Sequential Analysis, Fuzzy Cognitive Map, Scenario Modelling, Time Series.

Abstract: The paper considers the model of monitoring conducted to detect structural shifts in socio-economic processes,

arising under the influence of events in the external environment. The proposed procedure includes

monitoring of the macro- and business environment in which the observed process develops; monitoring of

time series describing the dynamics of the observed processes; supervisor for analysing signals from two

monitoring systems, adjusting their parameters and forming the final output signal. The practical significance

of the proposed procedure consists in increasing the efficiency of structural shift detection algorithms by

obtaining additional information by them, and, accordingly, in enhancing the capabilities of expert-analysts

and forecasters in solving the target problems of analysis and forecasting in situations of uncertainty and

instability based on the processing of heterogeneous information.

1 INTRODUCTION

In control methodology, monitoring is one of the

universal functions, necessary and applicable to any

control object (technical, socio-economic,

environmental, etc.). The relevance of this function is

due, firstly, to the increased requirements for the

control of objects of different nature, which are

constantly becoming more complex, and secondly, to

the need to account for rapid changes in the external

environment, affecting the control object, the

inconsistency of these changes and their interrelated

nature. (Rychihina, 2008, Aita, 2020)

The description of the processes of functioning of

social, financial and economic systems includes a set

of time-varying numerical parameters. There is a

need for timely detection of changes in the structure

of process development under the influence of

internal and external influences. Digital monitoring

systems conducted to detect changes in time series of

parameters called structural shifts have become

widespread (e.g., Lazariv and Schmid, 2018,

Pergamenchtchikov, Tartakovsky and Spivak, 2022).

However, the results of digital monitoring do not

allow us to solve such tasks as: (i) forecasting the

https://orcid.org/0000-0002-4517-6750

https://orcid.org/0000-0002-5153-8578

https://orcid.org/0000-0001-7675-5192

further development of the situation that led to the

emergence of a structural shift and the signal of

digital monitoring; (ii) forecasting the situation that

may lead to a structural shift. To support the solution

of these tasks, along with digital monitoring, it is

necessary to develop situation monitoring tools based

on the processing and analysis of expert information

from heterogeneous sources in order to identify the

causes of structural shifts in the observed processes,

to gain an in-depth understanding of the situation,

which will help the forecasting system to build

appropriate forecasts. By situation monitoring, we

mean dynamic tracking and evaluation of significant

changes (events, phenomena) in a situation that

characterizes the interaction of the observed

process(es) in the functioning of the systems

mentioned above with heterogeneous environmental

factors affecting these processes. The emphasis on the

processing of qualitative information in situation

monitoring is due to the lack of quantitative data on

the processes of functioning of these systems in

conditions of rapid (turbulent) and poorly predictable

changes in the external environment.

In this research, we present a model for organizing

the joint situation and digital monitoring of socio-

428

Avdeeva, Z., Grebenuyk, E. and Kovriga, S.

Model for Monitoring of Socio-Economic Processes Using Fuzzy Cognitive Map and Algorithms for Detecting Structural Changes.

DOI: 10.5220/0011949900003612

In Proceedings of the 3rd International Symposium on Automation, Information and Computing (ISAIC 2022), pages 428-433

ISBN: 978-989-758-622-4; ISSN: 2975-9463

 2023 by SCITEPRESS – Science and Technology Publications, Lda. Under CC license (CC BY-NC-ND 4.0)

economic processes. The proposed monitoring

procedure implements the following objectives:

detection of structural shifts, identification of the

causes that have led or may lead to the emergence of

structural shifts, formation of scenarios for possible

development of the situation, assessment of the power

and duration of upcoming changes.

Situation monitoring is aimed at identifying

information about significant changes in the external

environment that affect the dynamics of the observed

process. Situation monitoring organizes the

identification of such information through regular

source tracking of heterogeneous (structured and

unstructured) information and expert knowledge. It

includes the detection and identification of events and

system-forming factors that affect the dynamics of the

observed process and the formation of signals about

the state of the current situation. We call the factors

system-forming, since in their unity they form a

system that reflects a holistic view of the situation in

the context of the problem being solved. The

formation of signals is carried out on the basis of

processing information about potentially significant

events and factors by constructing scenarios for the

development of the situation with an assessment of

the significance of the effects of these events on the

observed process.

Digital monitoring reveals structural shifts that

occur when changing the dynamics of the series of

individual quantitative indicators and / or changing

the relationships between them: changes in trend,

level, variance. The methods of digital monitoring

used by us are based on algorithms of sequential analysis

(Wald, 1947, Page, 1954) adapted for monitoring non-

stationary processes (Grebenyuk, 2020).

Signals from situation and digital monitoring

systems go to the supervisor for organizing information

exchange between them and transmitting the

generated signals to the system for solving target

tasks related to the analysis of the influence of these

events on the observed process and the prediction of

its state.

In accordance with the purpose of the paper, the

contribution of this research is as follows:

1. We proposed a monitoring organization model

that includes, in addition to monitoring the time series

describing the dynamics of the process, monitoring

the situation in the macro and business environment,

and a supervisor with the following functions: 1) –

formation of input data for each type of monitoring;

2) implementation of information interaction between

the two monitoring systems; 3) transfer of aggregated

signals to the input of the tasks to be solved for the

research and forecasting of the state of the process

under study.

2. To implement situation monitoring, we have

developed an algorithm for generating signals about

the state of the external environment, formed as a

result of analysing significant events and modelling

the situation on the fuzzy cognitive map (FCM)

(Dickerson and Kosko, 1994, Avdeeva, Kovriga and

Makarenko, 2016). The FCM has found wide

application in knowledge engineering for structuring

and representing causal knowledge, which allows

formalizing the knowledge domain in the form of

domain-oriented concepts (factors) and relationships

between them, especially when solving problems in

situations of uncertainty (Cheah et al., 2008). In

addition, FCM is a model for integrating private

knowledge of diversified experts into a holistic

generalized view of the situation, thereby increasing

the validity of the conclusions (results) obtained on

its basis. In this research, the FCM is a model of

causal influences of system-forming factors that

reflect the situation of interaction between the

observed object and the external environment.

3. We have developed an algorithm for monitoring

groups of time series, each of which describes distinct

aspects of the development of the situation. The

monitoring algorithm is an ensemble of sequential

analysis algorithms configured to detect changes of

the following type: trend change, variance change.

2 GENERAL DESCRIPTION

OF THE MODEL

FOR MONITORING OF SOCIO-

ECONOMIC PROCESSES

Figure 1 shows the control scheme of the process of

detection, identification and assessment of significant

changes in the observed processes in situation and

digital monitoring modes. Top of Figure 1 shows the

process of forming and structuring the information

necessary for the monitoring procedure.

To get a holistic view of the situation (interaction

of the observed object with the external

environment), we perform structuring and

formalization of the subject area in the form of the

FCM, in which we distribute system-forming factors

into groups (SF-groups).

SF-groups characterize the belonging of system-

forming factors to certain aspects of the situation. For

example, for commodity markets these are market

factors of the business environment related to supply

and demand, financial and economic indicators of the

external environment, factors of influence of state

regulators, etc.

Model for Monitoring of Socio-Economic Processes Using Fuzzy Cognitive Map and Algorithms for Detecting Structural Changes

429

3. Supervisor

of monitoring

4. Solution of the target problem

2. Digital monitoring

Expert knowledge

Forming and structuring

a quantitative database

based on the distribution

of time series by SF-groups

Qualitative database of

- observed process, etc;

- significant events

Search. Heterogeneous

sources of information

Detection of the structural shift

type (SS) in the observed process

Y in the SFi-Group at the time

of observation t

Quantitative

database

t - a moment of observation

Defining infosearch parameters.

Identification and estimation

of system-forming factors

of the FCM associated

with significant info-occasions,

based on analysis and scenario

modelling on the FCM

Processing of the identified structural shift S

or the generated signals S

other

= Signal 1 and/or 2

1. Situation monitoring

other

Forming a qualitative database

FCM of situation

with grouping

of factors

into SF-groups

Сollection and accumulation of data

Update moment of observation t

Processing the results

of joint monitoring

other

= Signal 1 and/or 2

Figure 1: The control scheme of detection, identification and assessment of changes in the observed processes.

To get a holistic view of the situation (interaction

of the observed object with the external

environment), we perform structuring and

formalization of the subject area in the form of the

FCM, in which we distribute system-forming factors

into groups (SF-groups). SF-groups characterize the

belonging of system-forming factors to certain

aspects of the situation. For example, for commodity

markets these are market factors of the business

environment related to supply and demand, financial

and economic indicators of the external environment,

factors of influence of state regulators, etc.

We form the line-up of the databases based on the

results of a survey of experts, analysis of diverse

information sources, and recommendations from the

FCM of the situation. The qualitative data base

receives information about the structure of

interactions between the observed process and related

processes, about expertly significant events and

factors that affect changes in these processes. The

quantitative database includes time series of

processes, distributed in SF-groups.

Monitoring realizes the following functions:

detection of structural shifts – changes in the time

series properties describing the process dynamics

(Block 2 on Figure 1); detection and identification of

changes in the state of macro- and business

environment (Block 1 on Figure 1). The monitoring

supervisor (Block 3) is intended for processing the

results of joint monitoring, their confirmation and

transfer of responses to the target problem solving

block (evaluation of the current situation,

identification of causal relationships between groups

of parameters, forecasting on different time horizons,

etc.) (Block 4).

We form assessments of the significance of the

event consequences occurring in the external

environment and signals of situation monitoring by

using the following tools: (i) a model for representing

causal influences between significant system-forming

factors, reflecting the influence of the macro- and

business environment on the observed process – the

FCM of situation; (ii) author methods of structural

analysis and modelling based on the FCM of situation

ISAIC 2022 - International Symposium on Automation, Information and Computing

430

(Avdeeva, Kovriga and Makarenko, 2016, Avdeeva,

Grebenyuk and Kovriga, 2020a).

Situation monitoring (Block 1) keeps track

significant events that may influence the dynamics of

processes for which a certain target task is being

solved (forecasting the target indicator, analysing the

dynamics of relationships, comparative analysis of

SF-groups activity, etc.). Based on the results of the

analysis of significant events 𝐼𝑛𝑓, Block 1 generates

and transmits following signals to the supervisor:

Signal 1 – a reversal of target indicator change

direction, Signal 2 – significance weights of SF-

groups, or 0 – no signal (Signal

other

=Signal 1 and\or

Signal 2 or 0. For each time series in which a change

is detected, Block 2 generates, at the moment of

detection, a signal 𝑆𝑆 of the following structure:

{moment of detection t}, {time series name, name of

the SF-group it belongs to} {type of change: level,

trend, volatility} {parameter value after detected

change} {list of series that are causal by Grainger for

this series}. The signal from block 2 is transmitted to

the supervisor input.

In block 1 of situation monitoring, if a potentially

significant event Inf is detected, which can cause a

structural shift SS, the corresponding signals S

Signal 1 and\or Signal 2 are generated. The supervisor

sends this information to block 4 to correct the

algorithm of the target problem solution. If the event

is not detected, then the supervisor sends the signal

to Block 4.

3 DIGITAL MONITORING

3.1 Monitoring Object and Basic

Algorithm

The objects of digital monitoring considered here is

the set of time series, distributed among groups of

system-forming factors of the situation (SF-groups).

These objects have the following features: (i) the

series of observed indicators are non-stationary

processes with a stochastic trend; (ii) time series

belonging to the same SF-group follow general

trends. Examples of such objects can serve as time

series of commodity market prices subject to joint

movement (Liu et al., 2022).

The constructed digital monitoring system

consists of a set of sequential analysis algorithms

(Page, 1954), which detect changes in the parameters

of the observed time series distribution, provided that

these values are accurately known before and after the

change. To monitor each individual time series, we

apply the algorithm proposed in (Nikiforov, 2000).

The algorithm solves the following problem. Let

, 1, 2,...

YRt∈= is an independent sequence of

observations F(X

, Ɵ

) with parameter Ɵ

, which

after an unknown and non-random time moment

> begins to change according to the distribution

law F(X

, Ɵ

) with parameter Ɵ

, m=1,2,…,r. It is

required to detect as soon as possible the fact of

change of parameter θ and determine the type of

change m.

When next observation arrives, the algorithm

calculates a pair (N,

), where N is the point in time at

which the algorithm signals the presence of changes,

is the type of change. In (Nikiforov, 2002), it is

proved that at exactly known parameters of signal

distribution before and after changes the algorithm is

optimal by criterion of minimum of maximum

average delay of detection with restrictions on

frequency of false alarms and probability of false

diagnostics.

Since we do not know the exact values of the

parameters after the change, we determine the range

of values of the monitored parameters according to

historical data: no changes, small changes, medium,

large. For each area and for each type of shift, we

assign the values of the parameter Ɵ

, which are

different for different time series.

3.2 Algorithms for Monitoring

Non-Stationary Processes

The algorithm (Nikiforov, 2000) accepts a random

independent digital sequence as input, and the

observed object is a non-stationary series

integrated of the first order, so the problem arises of

generating signals supplied to the input of the basic

algorithm. To detect changes in non-stationary series,

we build an autoregression model based on the

differences

1ttt

YYY

−

Δ= − in an interval without

structural shifts:

titit

μβ

−

Δ=+ Δ +



)

where ( 1,..., 1)

=−are the coefficients of the

model;

is drift;

(0, )



is a random sequence.

We calculate its residuals, which we send to input on

the algorithm for detection changes

Re .

tt iti

μβ

−

=Δ − − Δ





)

The algorithm for detecting structural shifts of a

non-stationary series is proposed in [7] and includes

2 algorithms for detecting changes in drift and

variance along the vector of residuals (Eq. 2) of the

Model for Monitoring of Socio-Economic Processes Using Fuzzy Cognitive Map and Algorithms for Detecting Structural Changes

431

model (Eq. 1). When receiving a request from the

situation monitoring unit, the algorithm narrows the

range of permissible changes in tuning parameters in

order to increase the probability of detection due to

some increase in false alarms.

4 SITUATION MONITORING

The purpose of situation monitoring (Block 1 in

Figure 1) is (i) tracking environment events that affect

the dynamics of the observed process Y , which

cannot be identified by digital monitoring in

quantitative data (time series) at the time of

observation; (ii) the identification, assessment of the

significance of these events for changing Y and the

formation of appropriate signals for changing settings

or/and expanding the composition of objects of

observation in Block 2 of digital monitoring.

The basis for tracking and filtering information is

the FCM of the situation, which reflects the causal

influences between significant system-forming

factors that characterize the interaction of observed

process and the processes of the macro- and business

environment. The FCM with the distribution of

factors by SF-groups structures the knowledge

subject domain, which allows you to organize a

directed search for information in heterogeneous

sources.

The detected potentially significant info-occasion

Inf is associated with the system-forming factors

inf

{}



corresponding to it and evaluated on the basis

of modelling on the FCM S

inf

scenario of the

development of the situation “what will happen if”,

the output of which is the value of factor



characterizing the qualitative dynamics of the

observed process

Y and allowing us to assess whether

Inf affects the change in Y. We evaluate the modelling

results of the scenario S

inf

on the basis of the

developed estimates characterizing the significance

of the factor influence and their activity on the object

𝑌 dynamics of the digital monitoring (Avdeeva,

Grebenyuk and Kovriga, 2020b). We calculate these

estimates by the values of the integral influences

between any factors of FCM. Each such integral

assessment takes into account all direct and indirect

influences between a pair of factors in the FCM of

situation. According to the results of the scenario S

inf

evaluation, which reflects the influence of the event

Inf on the dynamics of Y

, we form signals S=Signal 1

and\or S=Signal 2 or S=0

( no signal). Signal 1

indicates a reversal of change direction of indicator

𝑦

due to event Inf. Signal 2 transmits the significance

weights (influence) of SF-groups on Y whose system-

forming factors were included in the scenario S

inf

These weights characterize the contribution of each

SF-group to the change in Y, taking into account the

significance of the influence of active factors

inf

{}



belonging to these groups on Y and the degree of

activity of manifestation of these factors in the

scenario S

inf

. The signal 𝑆=0 indicates the absence

of significant events affecting Y at the time of

observation

𝑡.

After processing, the supervisor (Block 3) sends

situation monitoring signals to Block 2 of digital

monitoring and to the target problem solving block

(Block 4) (Figure 1).

5 CONCLUSIONS

In modern conditions of instability of the

environment, the uncertainty of its impact on

changing the processes of functioning social,

economic and financial systems, the role of

monitoring the processes, the dynamics of which is

affected by the state of the external environment, is

increasing. In such systems, in addition to the

traditional monitoring of digital indicators, situation

monitoring based on the processing of expert

information is a necessary component of supporting

the solution of the target tasks of the analysis of

dynamic systems at different time horizons.

This paper presents a model of joint application of

situation and digital monitoring, which expands the

possibilities of digital monitoring by providing

additional information to it. The joint monitoring

algorithm includes: 1) digital monitoring algorithms

to detect structural shifts that use signals from

situation monitoring, in addition to observations of

digital indicators; 2) situation monitoring of observed

processes in interaction with the external

environment, conducted on the basis of processing

expert knowledge, tracing information from

heterogeneous sources about potentially significant

events of the external environment, scenario analysis

and modelling of the situation in order to assess the

expected consequences of these events (the

occurrence of structural shifts) on the dynamics of the

observed process.

We tested the efficiency of the proposed procedure

on the example of commodity market monitoring using

(1) information about environmental events and

significant factors affecting prices, (2) time series of

macroeconomic indicators, prices for metal products

and raw materials. We carried out the monitoring as

ISAIC 2022 - International Symposium on Automation, Information and Computing

432

part of solving the problem of correction on the

forecasting horizon of the constructed monthly forecast

of prices for ferrous scrap for 2019 (Avdeeva,

Grebenyuk and Kovriga (2021)). The experiment

showed that the forecast error is reduced by several

times (in comparison with the “naive” forecast) due to

the structuring of the situation, the formation of

forecasts using ensembles of models, the correction of

the situation on the forecast horizon based on the

results of situational monitoring and digital

monitoring. The experiment confirms that joint

monitoring improves the quality of detection of

structural shifts by digital monitoring due to the

information provided by situational monitoring, helps

to identify the causes of their occurrence and take this

information into account when forming a forecast in

order to improve its accuracy.

The practical significance of the proposed

monitoring procedure consists in increasing the

efficiency of structural shift detection algorithms by

obtaining additional information by them, and,

accordingly, in enhancing the capabilities of expert-

analysts and forecasters in solving the target problems

of analysis and forecasting in situations of uncertainty

and instability based on the processing of

heterogeneous information.

ACKNOWLEDGEMENTS

The results presented in the paper is partitionally

supported by grant of RSF № 23-21-00455,

https://rscf.ru/en/project/23-21-00455/.

REFERENCES

Aita, R., 2020. Which Complexity Characteristics Do

Economical Industries Present? International Journal

of Design & Nature and Ecodynamics, 15(2): 177-182.

https://doi.org/10.18280/ijdne.150206.

Avdeeva, Z., Kovriga, S., and Makarenko, D., 2016. On the

statement of a system development control problem

with use of SWOT-analysis on the cognitive model of

a situation. IFAC-Papers Online, 49(12):1838-1843.

Avdeeva, Z., Grebenyuk, E., and Kovriga, S., 2020a. The

Technology of the Strategic Goal-Setting and

Monitoring of a Manufacturing System Development

on the Basis of Cognitive Mapping. Chapter 3 in

Manufacturing Systems: Recent Progress and Future

Directions, edited by M.A. Mellal. Nova Science

Publishers, New York.

Avdeeva, Z.K., Grebenyuk, E.A., and Kovriga, S.V., 2020b.

Forecasting of Key Indicators of the Manufacturing

System in Changing External Environment. IFAC-

PapersOnLine, 53(2):10720-10725.

Avdeeva, Z.K., Grebenyuk, E.A., and Kovriga, S.V., 2021.

Construction of Multi-step Price Forecasts in

Commodity Markets Based on Qualitative and

Quantitative Data Analysis Methods. In: Dolgui, A., et

al. (eds) Advances in Production Management Systems.

Artificial Intelligence for Sustainable and Resilient

Production Systems. APMS 2021. IFIP Advances in

Information and Communication Technology, vol 630.

Springer, Cham. https://doi.org/10.1007/978-3-030-

85874-2_68.

Cheah, Wooi-Ping, Kim, Kyoung-Yun, Yang, Hyung-

Jeong, Kim, Soo-Hyung, and Kim, Jeong-Sik, 2008.

Fuzzy Cognitive Map and Bayesian Belief Network for

Causal Knowledge Engineering: A Comparative Study.

The KIPS Transactions: Part B, 15B(2):147-158.

Dickerson, J.A., Kosko, B., 1994. Virtual Worlds as Fuzzy

Cognitive Maps. Teleoperators and Virtual

Environments, 3(2):173-189.

Grebenyuk, E.A., 2020. Monitoring and identification of

structural shifts in processes with a unit root. In 13th

International Conference Management of large-scale

system development. https://doi.org/10.1109/

MLSD49919.2020.9247829.

Lazariv, T., Schmid, W., 2018. Challenges in Monitoring

Non-stationary Time Series. In: Knoth, S., Schmid, W.

(eds) Frontiers in Statistical Quality Control 12.

Springer, Cham. https://doi.org/10.1007/978-3-319-

75295-2_14.

Liu, C., Sun, X., Wang, J., Li, J., Chen, J., 2021. Multiscale

information transmission between commodity markets:

An EMD-based transfer entropy network. Research in

International Business and Finance, 55(1): 101318.

https://doi.org/10.1016/j.ribaf.2020.101318.

Nikiforov, I.V., 2000. A simple recursive algorithm for

diagnosis of abrupt changes in random signals. IEEE

Transactions on Information Theory, 4(7): 2740-2746.

Nikiforov, I.V., 2002. Optimal sequential change detection

and isolation|. In 15th Triennial World Congress of

IFAC, 35(7):29-34.

Page, E.S., 1954. Continuous inspection schemes.

Biometrika, 41(1-2):100–115.

Pergamenchtchikov, S.M., Tartakovsky, A.G., and Spivak,

V.S., 2022. Minimax and pointwise sequential

changepoint detection and identification for general

stochastic models. Journal of Multivariate Analysis,

190: 104977.

Rychihina, E.N., 2008. Monitoring role in the formation of

long-term plan of socio-economic development of the

municipality. Regional economy, 1(13). Article

number: 1303. URL: https://eee-region.ru/article/1303/

(In Russia).

Wald, A., 1947. Sequential Analysis. John Wiley & Sons,

Inc., New York.

Model for Monitoring of Socio-Economic Processes Using Fuzzy Cognitive Map and Algorithms for Detecting Structural Changes

433