Fuzzy cognitive map of pre-emergency prediction M Yu Micheev1, O V Prokofiev1 and A E Savochkin1 1 Department of information technologies and systems, Penza State Technological University, 1a/11 pr. Baidukova / ul. Gagarina, Penza, 440039, Russian Federation Abstract. The state of a technically complex object (TCO) during its operation is determined by means of a set of parametric sensors making it possible to determine dynamics features of both the external production environment and object operation parameters. The sensors output signals can display the presence of short-run, step-type and other loads, changes in the struc- ture of the signal random component, and also they make it possible to forecast the approach of the operating mode to the pre-emergency situation. The monitoring system development is possible on a conceptual basis of fuzzy cognitive maps. This approach allows not to make the mathematical tools used complex and to simulate the TCO control process. The system ap- proach realization method to designing in terms of fuzzy cognitive maps allows to develop scenarios of monitoring and control system operation, to evaluate the information value of conclusions based on the results of individual sensors signals processing. 1. Introduction A technically complex object (TCO) pre-starting procedure is connected with the analysis of readings of parametrical sensors performing the measuring conversion of physical quantities of different nature. The influence of operation objective conditions, the environment, the adjustment of task, any of which can result in a pre-emergency situation, reveals itself during preparation for the operating mode and during the object movement. Effects of temperature and time drift of sensors characteristics together with equipment wear, electromagnetic noise, vibrations, changes in the external physical environment, operational adjustment, revision of the TCO control objective may appear in the sensors output signals during the work [1]. System approach to TCO designing makes it possible to bring the incoming in- formation flows into a single system, to set priorities or the degree of an indicator significance, to de- termine cause-effect relationships and to draw final conclusions for the management of an object. Elements of the system here are not only physically existing sensor output signals that have passed analog-to-digital conversion, but also virtual components: the results of checking the properties of sig- nals in program modules, the results of logical rules execution. Despite of some elements virtuality the system approach can be fully implemented in the TCO control system model. TCO control system is a stable structure with specifically defined relationships between the elements and the state of the sys- tem can be assessed. The factor of building the control system serves to provide the ability to control the object both in decision support mode and in automatic mode. 2. Research objective and its practical significance The subject area of the research has been to control the state of the moving object in difficult running conditions. The possibility of fast changes of operating regime, structural vibrations, anomalous detec- tor’s readouts, which are caused by obstacles and collisions, has been taken into account. 502 The set of readouts of acceleration transducer, linear velocity sensor, angular velocity transmitter and shock pick-up has submitted the information for drawing up a logical conclusion on the current state of the moving object. The concept of object control system on the basis of fuzzy cognitive map has enabled to apply the flexible control approach and to choose the solution from the vast area of fea- sible solutions. The goal of the work is to build a model reflecting the process of TCO work monitoring with the ability to predict transition to pre-emergency condition because of a number of reasons or because of a combination of reasons. To achieve the goal we propose a model of monitoring system for TCO condition, operating on short time segments while controlling motion process and during preparatory stage. The implementation tool of the model is a fuzzy Cognitive Map [2] (Fuzzy Cognitive Map, FCM), which meets the research objective and system approach to solving problems. The elements of the control system are the factors of FCM represented as vertices of an oriented graph. The factors are characterized by a numeric level and are connected by arcs of the oriented graph, the arcs weights re- flect the causal relationship intensity. The factors that do not depend on other ones (within the control system under consideration) refer to the state of the external environment and other circumstances which cannot be influenced on during the TCO operating. The research experience has made it possible to improve the fundamentals of using fuzzy numbers and conclusions [3, 4] and to obtain self-adjusting maps with improved accuracy [5]. There appeared research area using fuzzy logic with indefinite relations (neutrosophic logic [6]). Methods for reducing FCM complexity [7] and methods for generating FCM models based on evolutionary genetic algo- rithms [8, 9, 10] are proposed. The practice of FCM applying is constantly enriched with new examples: production of biofuels, food and pharmaceutical products [11]; making political and engineering decisions [12]; modeling of public opinion formation by the mass media [13]; time series forecasting [14]. There appeared scien- tific works on the dynamic assessment of complex system risks taking into account the priority of the factors [15]. The prospects of applying the FCM toolkit proved by a variety of publications is simulta- neously combined with deficiency of ready-made solutions for creation and operation of engineering systems, in particular, systems for TCO controlling. So, there appeared the necessity to develop own FCM model and to verify with its help the possibility in principle to control TCO having many dy- namic state parameters. Tasks for the safe operation of TCO and the sequence of procedures for processing and analyzing data are depicted in the form of a diagram in Figure 1. Security policies require well-timed prediction of the TCO pre-emergency condition. Critical changes in the mode of the object operation can occur on a short interval of time series. The procedures for analyzing short time series should be fully automated, in order for the violation of the structure of the time series to be detected in "real time". Procedures for assessing the adequacy and accuracy of the time series model can be standard. The prediction of a possible TCO pre- emergency condition should be presented in a user-friendly form which easy-to-use as a component of decision support system. 3. Methodology The functional stability and operability of the control system using FCM mathematical apparatus have been researched. One of the approaches to constructing a model of relations development scenarios is the construc- tion of a fuzzy cognitive map, a variant of which is shown in Figure 1. Causal dependence in the form of arcs of the oriented graph from left to right reflects the logical chains: "change in the structure of the sensor output signal" - "test result" - "output". The arcs of graph directed from right to left corre- spond to the "subjective" part of the cause-effect relationships, they depend on the adjustment of the TCO operation mode, when the influence of the corresponding causative factors decreases signifi- cantly. The factors included in the model are characterized by initial level, the change of which is possible during the functioning of the system, as well as by relations with other factors with different intensity 503 of links. In addition, the model contains exogenous concepts that have no reason to change within the simulated system, but have such reasons in the external environment of TCO applying. The scheme shows that the factors e1 to e4 refer mainly to real external reasons, and e5 to e7 refer to "subjective" reasons that can be adjusted by changing the TCO's operating mode. The feedback in the system shown between e22 and e13, e14, e15 demonstrates the possibility of the adjustment. e1 Short-term external e8 Irwin test impact e16 Increasing absolute e9 Goldfeld–Quandt Presence of anoma- e2 noise values test, Spearman test lous values of time series e10 e17 Nonlinear growth of Glejser test e3 absolute noise values Heteroscedasity of deviations e11 e4 Increase in noise power Park test, White test e18 e21 (dispersion of fluctua- Lack of stability of Preemer- tion) time series gencies e12 Step change at the Assessment of trend e5 sensor output model quality e19 e13 Additive shift of Changing the exter- Trend shift detection trend e6 nal environment test e20 e14 Change of trend Changing the control Gujarati test e7 goal slope e22 e15 Chow test Adjusting operating mode Figure 1. Fuzzy cognitive map. The modeling is performed with the help of the method that is characterized by simplicity of im- plementation, the one that was used by Guillermo Ochoa de Aspuru in his Java application [16]: (k) 1 n (k 1) Li   Lij , n j 1 (k 1) Lij  Eij L j  (k 1) (k 1)  Li  I ij 100, i  1,n j  1,n. k is an iteration number; (k) Lij - level of factor i in the range from 0 to 100; (k 1) Lij - result of factor j influence on factor i; 504 E - direction of influence, which takes on values -1 or 1; ij I - the intensity of causal connection, which takes on a value in the range from 0 to 100. ij 4. Technology of the experiment and the results obtained The instrument of the experiment was the Java-based modeling application which can be found on Guillermo Ochoa de Aspuru’s website [16] and which functions in the Internet Explorer environment. Levels of factors and intensity of effects are estimated in conventional units in the range from 0 to 100 and appear as a result of processing experts’ opinions. The values of the intensity of effects are presented in Table 1. Table 1. The intensity of effects. Dependent factor Causal factor Intensity of effect 75 e16 100 75 e17 100 75 e21 e18 100 75 e19 100 75 e20 100 75 e22 e21 100 other effects 50 As follows from Table 1, several variants of the intensity of effects with the factors-causes are con- sidered for the factors e21 and e22 in the experiment. The inventory of factors is created in the main dialog box of the program, and the values of the intensities of the relationships (effects) and the initial values of the factor levels are input in the editor window. The iterative process of the map transition to a new stable state stopped in accordance with the zero norm of correction between state vectors of the factors at neighboring iterations (run until conver- gence). The research results using a cognitive map have been obtained as iteration processes results (Con- vergence reached), the processes convergence has been achieved in all cases. 5. Analysis of the experiments results The final states of the factors of Fuzzy cognitive map are shown in Table 2. The columns correspond to the experiments numbers. Table 2. Examples of resulting factors states. 1 2 3 4 5 6 7 е1 0 0 0 0 100 0 100 е2 0 0 0 0 100 0 100 505 е3 0 0 0 0 100 0 100 е4 0 50 50 100 100 0 100 е5 0 0 0 0 100 100 0 е6 0 0 0 0 100 100 0 е7 0 0 0 0 100 100 0 е8 0 0 0 0 100 0 100 е9 0 0 0 0 100 0 100 е10 0 0 0 0 100 0 100 е11 0 50 50 100 100 0 100 е12 0 0 0 0 100 100 0 е13 0 0 0 0 100 100 0 е14 0 0 0 0 100 100 0 е15 0 0 0 0 100 100 0 е16 0 0 0 0 100 0 100 е17 0 16 16 33 100 0 100 е18 0 0 0 0 100 100 0 е19 0 0 0 0 100 100 0 е20 0 0 0 0 100 100 0 е21 0 3 3 6 100 60 40 е22 0 3 3 6 100 60 40 К 2 8 8 8 12 12 11 The column numbers correspond to the experiments numbers that are listed below. 1. Zero states of factors mean the system elements operation in the normative state and the absence of conclusion about a pre-emergency situation. The iterative process consists of two iterations. 2. Noise power increase (deviation variance) is represented by e4 equal to 50, and this was revealed with the help of the Park and White tests (e11). The pre-emergency situation factor e21 has level 3. It is recommended to increase the intensity of cause-effect relationships. The iterative process consists of 8 iterations. 3. The intensity of effect between e21 and concepts-causes equal to 75, between e22 and e21 equal to 75, with other unchanged conditions. The concept level of the pre-emergency situation e21 has not changed. The pre-emergency situation factor e21 has level 3. The iterative process consists of 8 itera- tions. 4. Noise power increase (deviation variance) is represented by e4 factor, its status is estimated as 100. The state of pre-emergency factor e21 has increased to 6. The iterative process consists of 8 itera- tions. 5. All the factor-causes from e1 to e7 are active and have the value of 100. The intensity of effect between e21 and the factor-causes is 100, between e22 and e21 is 100, with other conditions unchanged. All other factors reach the level of 100. The iterative process is completed within 12 iterations. 6. The only factor-reasons that are active are those from e5 to e7 and they have the value of 100. A pre-emergency situation reasons are related with the task of TCO moving, it is possible to adjust them. 506 The factor-causes from e1 to e4 have the value 0. The state of pre-emergency situation factor e21 has reached 60. The iterative process is completed within 12 iterations. 7. The situation that is reverse to the previous one. The only factor-causes that are active are those from e1 to e4 and they have the value of 100. The factor-reasons from e5 to e7 have the value 0. The state of pre-emergency situation factor e21 has changed to 40. The iterative process is completed within 11 iterations. 6. Practical significance The authors carry out research and development in predicting pre-emergency situations for technically complex objects. Produced fuzzy cognitive map (Figure 1) allowed to develop the conception of the functioning complex computer system (CCS) prediction of emergency on the basis of analysis of structural insta- bilities and inhomogeneities of time series [17]. This determines CCS purpose, tasks, functions and principles of functioning, which allows defining the directions of works on its creation, use and devel- opment. Figure 2. Conception of the functioning construction of CCS prediction of emergency situations. In the process of developing CCS for forecasting an emergency situation on the basis of an analysis of time series [18], the following principles should be guided: 507 - parallelism (provides reduction of processing time, collection and analysis of initial information and execution of the very definition of the possibility of an emergency); - continuity (the principle that operates after the very process of determining the possibility of an emergency situation and making corrections, as necessary, ensuring systematic collection and process- ing of additional information) [19]; - directness (reflects a strictly expedient, regulated transfer of information on the shortest path); Within the framework of such CCS, in order to forecast an emergency, time series should be con- ducted through several stages of analysis: - determination of the structural instability of time series and the search for anomalies in it [20-22]; - determination of the presence of a trend in time series [23-26]; - study of the change in the dispersion of the residual component of the time series (revealing het- eroscedasticity). At each of the stages considered, an assumption is made about a future change in the process. At the end of all stages, a general assumption is made about the occurrence of an emergency. 7. Conclusions Within the framework of the application of fuzzy cognitive map as unified formal model of forecast- ing the occurrence of an emergency on the basis of the analysis of time series, the following elements are consolidated: - discrete mathematical models of time series; - a complex of mathematical models of structural inhomogeneities of time series; - a complex of mathematical models of structural instabilities of time series; - conditions for the optimal application of each model; - generalized conception of TCO emergency situation prediction based on the analysis of the time series structure. Experiments 2-4 prove the importance of adjusting the control system sensitivity both via the fac- tors levels and intensities of cause-effect relations. It is disputable that there is the relation between the factors levels of the pre-emergency factor e21 having values 60 and 40 corresponding to experi- ments 6 and 7. Despite this the testing of the model demonstrates its adequacy in a wide sense, since there are no discrepancies between finite steady states of the system and fundamental concepts in the subject area. To achieve functional stability and efficiency of the monitoring system is possible only with the complex approach to threat evaluation of approaching to the pre-emergency situation. Fuzzy cognitive maps tools used here are rather fruitful in terms of covering related subject areas [6]. It is not associ- ated with a significant increase in the complexity of the algorithm with the complexity of parametric sensor equipment. 8. References [1] Prokofev O, Savochkin A Modeling of the detection system of the preliminary situation based on the fuzzy cognitive map. Reliability and quality of complex systems. vol 2 (22), 2018, pp 73- 79. DOI: 10.21685/ 2307-4205-2018-2-10 [2] Kosko B 1993 Fuzzy Thinking: The New Science of Fuzzy Logic. Hyperion, New York, p 320 [3] Froelich W, Salmeron J L 2017 Advances in fuzzy cognitive maps theory. Neurocomputing. vol 232, 5 April 2017 pp 1-2. https://doi.org/10.1016/j.neucom.2016.11.058 [4] Dodurkaab M F, Yesilab E, Urbasa L 2017 Causal effect analysis for fuzzy cognitive maps de signed with non-singleton fuzzy numbers. Neurocomputing vol 232, 5 April 2017 pp 122-132 https://doi.org/10.1016/j.neucom.2016.09.112 [5] Štula M, Marasa J, Mladenović S 2017 Continuously self-adjusting fuzzy cognitive map with semi-autonomous concepts. Neurocomputing vol 232, 5 April 2017 pp 34-51. https://doi.org/10.1016/j.neucom.2016.09.114 [6] Vasantha Kandasamy W B, Smarandache F 2003 Fuzzy Cognitive Maps and Neutrosophic Cogni tive Maps. Xiquan, Phoenix p 212 508 [7] Papageorgiou E I, Hatwágner M F, Buruzsc A, Kóczy L T 2017 A concept reduction approach for fuzzy cognitive map models in decision making and management. Neurocomputing vol 232, 5 April 2017 pp 16-33. https://doi.org/10.1016/j.neucom.2016.11.060 [8] Stach W, Kurgan L, Pedrycz W, Reformat M 2005 Genetic learning of fuzzy cognitive maps. Fuzzy Sets and Systems vol 153, Issue 3, 1 August 2005 pp 371-401 https://doi.org/10.1016/j.fss.2005.01.009 [9] Karel Mls, Cimler R, Vaščák J, Puheim M 2017 Interactive evolutionary optimization of fuzzy cognitive maps. Neurocomputing. vol 232, 5 April 2017 pp 58-68 https://doi.org/10.1016/j.neucom.2016.10.068 [10] Chena Ye, Mazlack L, Minai A, Long J Lu 2015 Inferring causal networks using fuzzy cogni- tive maps and evolutionary algorithms with application to gene regulatory network recon- struction. Applied Soft Computing vol 37 (2015) pp 667–679. https://doi.org/10.1016/j.asoc.2015.08.039 [11] Salmeron J, Ruiz-Celma A, Menae A 2017 Learning FCMs with multi-local and balanced memetic algorithms for forecasting industrial drying processes. Neurocomputing vol 232, 5 April 2017 pp 52-57 https://doi.org/10.1016/j.neucom.2016.10.070 [12] Christoforou A, Andreou A 2017 A framework for static and dynamic analysis of multi-layer fuzzy cognitive maps. Neurocomputing vol 232 5 April 2017 pp 133-145. https://doi.org/10.1016/j.neucom.2016.09.115 [13] Aguilar J, Téran O, Sánchez H, Gutiérrez de Mesa J, Cordero J, Chávez D 2017 Towards a Fuzzy Cognitive Map for Opinion Mining. Procedia Computer Science vol 108 2017 pp 2522-2526. https://doi.org/10.1016/j.procs.2017.05.287 [14] Homenda Wl, Jastrzebska A 2017 Clustering techniques for Fuzzy Cognitive Map design for time series modeling. Neurocomputing vol 232 5 April 2017 pp 3-15. https://doi.org/10.1016/j.neucom.2016.08.119 [15] Jamshidi A, Daoud Ait-kadi, Ruiz A, M Larbi Rebaiaia 2018 Dynamic risk assessment of com- plex systems using FCM International Journal of Production Research. vol 56 2018 Issue 3 pp 1070- 1088. https://doi.org/10.1080/00207543.2017.1370148 [16] Guillermo Ochoa de Aspuru Fuzzy Cognitive Maps http://www.ochoadeaspuru.com/fuzcogmap/index.php [17] Shao X, Zhang X 2010 Testing for Change Points in Time Series. Journal of the American Statisti cal Association September 2010 vol 105 491 Theory and Methods pp 1228-1240. [18] Andrews D 2003 End-of-Sample Instability Tests. Econometrica vol 71 6 November 2003 pp 1661-1694. [19] Badagián A, Kaiser R, Peña D 2013 Time series segmentation procedures to detect, locate and es timate change-points. Empirical economic and financial research. Advanced studies in theoreti- cal and applied econometrics vol 48 May 2013 p 174. [20] Irwin J 1925On a Criterion for the Rejection of Outlying Observations. Biometrika vol 17 3 4 1925 pp 238-250. [21] Chow G 1960 Tests of Equality Between Sets of Coefficients in Two Linear Regressions. Econometrica 1960 vol 28 (3) pp 591–605. [22] Gujarati D 2011 Econometrics by Example Palgrave Macmillan New York 2011 p 371. [23] Hinich M J, Foster J, Wild P 2006 Structural change in macroeconomic time series: a complex sys tems perspective. Journal of Macroeconomics vol 28 1 March 2006 pp 136–150. [25] Jouini J, Boutahar M 2005 Evidence on structural changes in U.S. time series. Economic Model- ling vol 22 2005 pp 391–422. [26] Chen B, Hong Y M 2012Testing for smooth structural changes in time series models via nonpara metric regression. Econometrica vol 80 3 May 2012 pp 1157-1183. 509