CC BY
48
doi 10.24411/2409-5419-2018-10045
EMPLOYMENT OF FUZZY NEURAL NETWORKS FORECASTING PROFESSIONAL SUCCESS ACTIVITIES OF THE MILITARY EXPERTS
PETRICH
Dmitriy Olegovich1
Okhotnikov Yuriy Yur'yevich2
SHAYMUKHAMETOV Shamil' Il'dusovich3
Information about authors:
1PhD, Senior Lecturer of the Military Space Academy, St. Petersburg, Russian, [email protected];
2Lecturer of the Military Space Academy, St. Petersburg, Russian, [email protected];
3Postgraduate Student of the Military Space Academy, St. Petersburg, Russian, 28 [email protected]
ABSTRACT
In the modern world, the ability of a specialist to adapt to the dynamically changing conditions of his professional activity becomes very important. This task is very important in the training of highly qualified personnel in the higher military educational establishments of the Ministry of Defense of the Russian Federation. The high cost of training qualified military specialists, the high level of requirements imposed on the results of their professional activities, makes it extremely important to solve the problem of forecasting and early evaluation of the success of further professional activities of graduates of higher educational institutions of the Ministry of Defense of the Russian Federation. The success of the professional activity of a graduate is determined by the correspondence of professionally important qualities to the requirements for his future military professional activity.
The most preferred mathematical apparatus for modeling such a class of problems, where there are a lot of indistinctly expressed input data, in the aggregates of which the laws and interrelations between them are hidden, is the apparatus of odd neural networks. The expediency of using fuzzy neural networks is also conditioned by incomplete or indistinct information of preferences, as well as by intuitively formulated rules for solving such problems. To implement the process of evaluating performance and predicting the success of a graduate, it is proposed to consider the class of adaptive networks functionally equivalent to systems of fuzzy reasoning. Such an architecture is called ANFIS. ANFIS is one of the first variants of hybrid neural-fuzzy networks - a neural network of direct signal propagation of a special type. The architecture of the neural-fuzzy network is isomorphic to the fuzzy knowledge base. In neural-fuzzy networks, differential implementations of triangular norms (multiplication and probabilistic OR), as well as smooth membership functions, are used. This allows us to apply fast neural network training algorithms based on the method of back-propagating the error to configure neural-fuzzy networks. ANFIS implements a fuzzy inference system in the form of a five-layer neural network of direct signal propagation.
The use of the proposed approach will help with the selection of the most appropriate and exclusive low-information methodologies for professional and psychological selection, with the selection of the most effective teaching methods. Different preferences can be justified using the methods of one-dimensional and multivariate statistics. After this, the development of an algorithm (a decisive rule) for evaluating occupational fitness is carried out. Most often for these purposes, use multiple regression analysis, based on the relationship of psychophysiological properties with "external criteria," which refers to the quality (success) of training or activity.
KEYWORDS: forecasting; professional activity; a graduate of the military school; fuzzy neural network; regression analysis.
For citation: Petrich D. O., Okhotnikov Yu. Yu., Shaymukhametov Sh. I. Employment of fuzzy neural networks forecasting professional success activities of the military experts. H&ES Research. 2017. Vol. 10. No. 2. Pp. 100-106. doi 10.24411/2409-5419-2018-10045
Vol 10 No 2-2018, H&ES RESEARCH PUBLICATIONS IN ENGLISH
INTRODUCTION
The ability of different specialists to adapt to the dynamically changing conditions of their professional activity becomes very important in the modern world. Also this problem is very important in the process of highly qualified personnel training in the higher military educational institutions of the Ministry of Defense (MoD) of the Russian Federation (RF). The high cost of training skilled military specialists, a high level of requirements for the performance of their professional activity, causes an extremely high importance of solving the problem of forecasting and early evaluation of success of future professional graduate's activity.
The success of the professional graduate's activity is determined by the relevant professional-important qualities that he possesses the requirements for his future military career. Experience shows that a person does not have the ability to certain professional activities which are not only much longer than the others, and with great difficulty seize this activity, but also often make mistakes and failures are to blame for accidents, accidents and emergency situations. Thus, early forecasting of the development path of a military specialist can help in the most effective organization of his future activities.
Approach to predicting the success of professional activity of military specialists on the basis of fuzzy neural networks
Professional activity of a military specialist is complex activity that appears to a specialist as a constituted way of doing something that has a normatively established character. Professional activity is objectively complex, so it is difficult to master, requires a long period of theoretical and practical training [5].
The success of the professional activity of a specialist is determined by his readiness for a certain type of professional activity.
Readiness for professional activity is a psychological state, pre-start activation of a person, including a person's comprehension of his goals, assessment of existing conditions, the definition of the most probable ways of action, predicting motivational, volitional, intellectual efforts, the probability of achieving results, mobilizing forces, self-hypnosis in achievement of goals [4].
Summarizing this definition, one can consider readiness for professional activity as a multilevel and multifaceted system-structural personal formation of the individual.
Most often, it is common to allocate such components of readiness for professional activity as motivational (positive attitude towards the future profession), orientational (knowledge of the profession), operational (professional thinking, set of skills), volitional (self-regulation and behavior management), evaluative (self-evaluation professional preparedness) [7, 13]. Accordingly, the assessment of a specialist's readiness for professional activity is a necessary action and consists in a
comprehensive evaluation of the system of integrative properties and qualities of the individual, as well as the knowledge, skills and skills of the specialist.
There are many different methods for identifying the most suitable specialists at the stage of vocational selection, assessing the quality of education and improving the final training and performance of graduates, which basically use methods of quantitative assessment, but assessing a specialist's readiness for professional work also implies an assessment of a number of qualitative characteristics that are often are of fuzzy nature, and also have different dimensions, meaning and contribution to the integrated indicator [12]. Thus, the problem of assessing the readiness of specialists for professional activity is reduced to the problem of classifying their states on the basis of a huge amount of initial data. The necessity to take into account the psychological, physiological characteristics of individual individuals, the conditions of their work as a result leads to the task of constructing a separating surface, described by a complex multicriteria function.
Proceeding from the above, it can be concluded that the questions of assessing the readiness of specialists for professional activity are a complex task that can be attributed to the tasks and recognition of objects belonging to overlapping classes. One of the most common ways to solve this class of problems is to use mathematical models of neural fuzzy production networks that connect the capabilities of fuzzy inference systems and neural networks [12].
The paper proposes an approach to forecasting the success of professional activities of graduates on the basis of a mathematical model of a fuzzy neural network.
This approach allows to link together all the stages of training and evaluation of a specialist, starting with the selection process, the training process and subsequent evaluation of the graduate's career [1]. The adaptive nature of the fuzzy neural network is also important. It is based on a fuzzy logical inference. In addition, it allows to reconfigure the parameters of the membership functions and to train the neural network.
The most preferred mathematical apparatus for modeling such class of problems, where there are a lot of indistinctly expressed input data, in the aggregate of which the laws and interrelations between them are hidden, is the apparatus of fuzzy neural networks. The expediency of using fuzzy neural networks is also due to incomplete or indistinct information of preferences, as well as intuitively formulated rules for solving such problems.
Psychological fitness for a profession is a property of a person, which can be judged by two criteria: successful mastery of the profession and the degree of satisfaction of a person with his labor. Both these criteria are relative, and sometimes subjective. Nevertheless, these criteria allow us to approach the characterization of professional suitability and subsequently evaluate the success of professional activity. At the very first
stage, the preliminary professional selection of cadets, taking into account the peculiarities of their future professional activity, is very important.
The basis for making an expert decision in the professional selection is the assessment of professional suitability. Profitability in selection is a likely characteristic reflecting a person's ability to master any professional activity. In the professional selection proficiency can be assessed by several criteria:
— on medical indicators, including on indicators of physical readiness;
— according to the educational qualification (results of the USE);
— with the help of psychological examination;
— taking into account some indicators reflecting the applicant's social status;
— taking into account the achieved level of professional adaptation, etc.
In this case, the forecast of the success of training and follow-up is based on the comparison of information about the requirements of the profession to a person and the psychodiagnostic data obtained, with an emphasis on the evaluation of personal characteristics; on the possibility of targeted improvement and compensation of professionally significant qualities; the likelihood of adaptation to the profession; the possibility of emergence of extreme effects.
Forecasting has a probabilistic evaluation and is based on the study of the structure of the personality, the structure of activity and on the correlation of these structures, which includes a very important component as the process of training the profession.
The second important component of forecasting is the "success of training"—an integral characteristic of the success of the training work. Most often, the success of training is assessed by performance indicators. The average score of training is used "as the most non-differentiated indicator of general abilities". For these purposes, the average score of training is used, and the average academic performance in the cycles of subjects over a long period of time. Assessments of the quality of training students are the results of ongoing monitoring of academic performance, intermediate and final attestation [1].
The third component of forecasting is based on information about the professional activities of graduates. This information is contained in the official responses to the graduates after the first year of their service in the troops. The service review provides an integrated assessment of training and performance through the use of multi-level scales of assessments for the most varied indicators, distributed across seven sections. The Academy is carrying out research to improve the methods of training military specialists, within the framework of which principles for the formation of a final assessment of the training and performance of graduates have been developed. A special computer program allows to automate the process of collecting, processing and analyzing data on submitted service reports [3].
Fuzzy neural networks draw conclusions based on the mathematical apparatus of fuzzy logic, but the parameters of the membership functions are tuned using neural network learning algorithms. Therefore, to select the parameters of such networks, we apply the method of back propagation of the error, originally proposed for training a multilayer percep-tron. For this, the fuzzy control module is represented in the form of a multilayer network. Fuzzy neural network, as a rule, consists of four layers: the layer of fuzzification of input variables, the layer of aggregation of activation values of the condition, the layer of aggregation of fuzzy rules and the output layer [8-9, 14-15].
Fuzzy associative rules are tool for extracting regularities from databases that are formulated in the form of linguistic utterances. Here are introduced special concepts of fuzzy transaction, support and reliability of fuzzy associative rules.
Fuzzy inference algorithms differ, mainly, by the kind of rules used, logical operations and the type of defuzzification method. Various models of fuzzy inference have been developed (Mamdani, Sugeno, Larsen, Tsukamoto).
Let us consider in more detail the fuzzy conclusion on the example of the Mamdani mechanism [11]. This is the most common method of inference in fuzzy systems. It uses the minimax composition of fuzzy sets. This mechanism includes the following sequence of actions:
1. The procedure of fuzzification: determine the degree of truth, i.e. the values of the membership functions for the left parts of each rule (prerequisites). For a rule base with m rules, the degrees of truth are denoted
A (X), i = 1,m, k = 1,n.
2. Fuzzy conclusion. First, the "cut-off" levels for the left side of each rule are determined
K = min (((x)).
Next, there are "truncated" membership functions
B't(y) = min((,B. (y)).
3. Composition, or combination of the obtained truncated functions, for which the maximum composition of fuzzy sets is used
MF (y ) = max. (B't(y)),
where is the membership function of the final fuzzy set.
4. Defuzzification, or reduction to clarity. There are several methods of defuzzification. For example, the method of the middle center, or the centroid method.
Based on the fuzzy logic inference algorithm, a system of reasoning is constructed (fig. 1).
Vol. 10. No. 2-2018, H&ES RESEARCH
PUBLICATIONS IN ENGLISH
Fig. 1. The system of fuzzy reasoning
The system of fuzzy reasoning consists of five functional blocks:
— a block of fuzzification that converts numerical input values to a degree of compliance with linguistic variables;
— a rule base containing a set of fuzzy rules such as "if something";
— a database in which the fuzzy set membership functions used in fuzzy rules are defined;
— decision-making unit that performs withdrawal operations on the basis of existing rules; — block of defuzzifi-cation, which converts the results of output into numerical values.
Traditionally, the rules database and the database are combined into a common block — the knowledge base.
Next we propose to consider the class of adaptive networks functionally equivalent to systems of fuzzy reasoning. This architecture is called ANFIS (an abbreviation Adaptive-Network-Based Fuzzy Inference System — adaptive network of fuzzy inference). ANFIS is one of the first variants of hybrid neural-fuzzy networks — a neural network of direct signal propagation of a special type. The architecture of the neural-fuzzy network is isomorphic to the fuzzy knowledge
base. In neural-fuzzy networks, differentiable implementations of triangular norms (multiplication and probabilistic OR) are used, as well as smooth membership functions. This allows us to apply fast neural network training algorithms based on the method of back propagation of the error to configure neural-fuzzy networks. The architecture and rules for the operation of each layer of the ANFIS network are described below. ANFIS implements a fuzzy inference system in the form of a five-layer neural network of direct signal propagation [6-9].
The purposes of the layers are:
— the first layer — the terms of the input variables;
— the second layer—antecedents (parcels) of fuzzy rules;
— the third layer — the normalization of the degree of implementation of the rules;
— the fourth layer — the conclusion of the rules;
— the fifth layer — the aggregation of the result obtained by different rules.
The network inputs in a separate layer are not allocated. Figure 2 shows an ANFIS network with two input variables X and x2) and four fuzzy rules. For linguistic evaluation of the input variable x1 3 terms are used, for a variable x2 2 terms are used.
Fig. 2. Network example
The ANFIS network works as follows [8]:
1. The 1st layer is the terms of the input variables. Each node of the first layer represents one term with the membership function. The number of nodes of the first layer is equal to the sum of the powers of the term-sets of the input variables. The output of the node is the degree to which the value of the input variable belongs to the corresponding fuzzy term. The parameters of this layer refer to the so-called prerequisites parameters.
2. The second layer is the antecedents of the fuzzy rules. Each node of a given layer is a fixed node multiplying input signals, with the output value of the node being the weight of a rule: The number of nodes of the second layer is m. Each node of this layer corresponds to one fuzzy rule. The node of the second layer is connected to those nodes of the first layer, which form the antecedents of the corresponding rule. Therefore, each node of the second layer can receive from 1 to n input signals. The output of the node is the degree of execution of the rule, which is calculated as the product of the input signals.
3. The third layer is the normalization of the degree of fulfillment of the rules. Each i-th node of this layer determines the ratio of the weight of the /-th rule to the sum of the weights of all rules: The output signals of the 3rd layer are called normalized weights. The number of nodes of the third layer is also equal to m. Each node in this layer calculates the relative degree of fuzzy rule execution.
4. 4th layer—the conclusion of the rules. The nodes of a given layer are defined by linear functions of the belonging of the output variables. The number of nodes of the fourth layer is also equal to m. Each node is connected to one node of the third layer, and also to all inputs.
5. The 5th layer is the aggregation of the result obtained according to different rules. The only node of this layer is a fixed node in which the total output value of the adaptive network Y is calculated as the sum of all input signals.
CONCLUSION
The use of the proposed approach will help with the selection of the most appropriate and the exclusion of little-informative methods of professional and psychological selection, with the selection of the most effective teaching methods. Different preferences can be justified using the methods of one-dimensional and multivariate statistics. After this, the development of an algorithm (a decisive rule) for evaluating occupational fitness is carried out. Most often for these purposes, use multiple regression analysis, based on the relationship of psychophysiological properties with "external criteria," which refers to the quality (success) of training or activity.
REFERENCES
1. Baibakov M. N., Bobrovskaya A.A. Prognozirovanie us-peshnosti professional'noy deyatel'nosti kursantov uchebnykh za-vedeniy GPS MChS Rossii na osnove matematicheskoy modeli
nechetkoy neyronnoy seti [Predicting the success of professional activity of cadets of educational institutions of State Fire Service of EMERCOM of Russia on the basis of mathematical model of fuzzy neural network]. Materialy mezhdunarodnoy nauch-no-prakticheskoy konferentsii "Podgotovka kadrov v sisteme pre-duprezhdeniya i likvidatsii posledstviy chrezvychaynykh situatsiy" [Materials of the International Scientific and Practical Conference "Training of personnel in the system of prevention and liquidation of consequences of emergency situations" (St. Petersburg, October 24, 2013)]. St. Petersburg: State Fire Service University of EMERCOM of Russia, 2013. Pp. 127-130. (In Russian)
2. Viktorova E. V. Application of fuzzy neural networks for technical diagnostics of road vehicles. Bulletin of Kharkov National Automobile and Highway University. 2012. Vol. 56. Pp. 98-102. (In Russian)
3. GolubevM.A., VoronkovI. Yu.,Mashkov O. G. Metodika otsenki udovletvorennosti zakazchika kachestvom podgotovki vypusknikov akademii na osnove analiza sluzhebnykh otzyvov [Methods of assessment of attorney-client satisfies the quality of training of graduates of the Academy on the basis of office reviews analysis]. Trudy Voenno-kosmicheskoy akademii imeni A. F. Mozhayskogo [Proceedings of the A. F. Mozhayskiy Military Space Academy]. 2014. Vol. 644. Pp. 207-211. (In Russian)
4. Gotovnost' k professional'noy deyatel'nosti. Slovar' po proforientatsii i psikhologicheskoy podderzhke [Readiness for professional activity. Dictionary on career counseling and psychological support]. URL: https://career_counseling_sup-port.academic.ru/75/Reportability_pro_professional_action. (date of access 01.10.2017).
5. Druzhilov S.A. Psihologija professionalizma chelove-ka: integrativnyj podhod [Psychology of human professionalism: an integrative approach]. Zhurnal prikladnoj psihologii [Journal of Applied Psychology]. 2003. No. 4-5. Pp. 35-42. (In Russian)
6. Dudkin A. A. Fuzzy neural network for the analysis of the topology of integrated microcircuits. Artificial Intelligence. 2015. No. 1-2. Pp. 79-86.
7. Zyryanova A. V Readiness for professional activity of specialists in the sphere of culture: essence and structure. Pedagogy of art. 2012. No. 4. URL: http://www.art-education. ru/AE-magazine/ № 4, 2012. (In Russian)
8. Ivaskiv Yu. L., Levchenko V V., Leshchinsky O. L. Formation of fuzzy learning sets for neural networks in problems of data compression without losses. Mathematical machines and systems. 2009. No. 2. Pp. 53-60.
9. Lubentsova E. V. Investigation of algorithms for learning the neuro-fuzzy control system of the biotechnologi-cal process. Scientific journal KubSAU. 2017. No. 128 (04). Pp. 1-11. (In Russian)
10. Matkovskaya M. O. Investigation of fuzzy inference algorithms in decision-making models. Izvestiya SFU. Technical Sciences. 2009. Pp. 240-243. (In Russian)
Vol 10 No 2-2018, H&ES RESEARCH INFORMATICS, COMPUTER ENGINEERING AND CONTROL
11. Melkov D. A. Sravnenie algoritmov nechetkogo vy-voda s ispol'zovaniem yazykov standarta MEK [Comparison of fuzzy inference algorithms using the IEC standard languages]. Young Scientist. 2013. No. 5. Pp. 74-79. (In Russian)
12. Mikhelkevich V. N., Kravtsov P. G. Comprehensive assessment of graduates' readiness for professional work. Samara Journal of Science. 2016. No. 2 (15). Pp. 171-175. (In Russian)
13. Pleshakova O. V Components of psychological readiness for the professional work of a social worker. Vestnik Bashkirskogo universiteta [Bulletin of Bashkir University]. 2007. Vol. 12. No. 3. Pp. 200-203. (In Russian)
14. Soldatova O. P., Lyozin I.A., Lyozina I. V, Kupri-yanov A. V., Kirsch D. V Application of fuzzy neural networks to determine the type of crystal lattices observed on nanoscale images. Computer Optics. 2015. Vol. 39. No. 5. Pp. 787-795. doi: 10.18287 / 0134-2452-2015-39-5-787-794 (In Russian)
15. Soldatova O. P., Lezin I.A. Solution of the classification problem using neural fuzzy production networks based on the Mamdani-Zade model of inference. Vestnik Samarsko-go Gosudarstvennogo Tekhnicheskogo Universiteta. Seriya Fiziko-Matematicheskie Nauki [Bulletin of the Samara State Technological University. Series Physics and mathematics]. 2014. No. 2 (35). Pp. 136-148. doi: 10.14498 / vsgtu1266 (In Russian)
ПРИМЕНЕНИЕ НЕЧЕТКИХ НЕЙРОННЫХ СЕТЕЙ В ПРОГНОЗИРОВАНИИ УСПЕШНОСТИ ПРОФЕССИНАЛЬНОЙ ДЕЯТЕЛЬНОСТИ ВОЕННЫХ СПЕЦИАЛИСТОВ
ПЕТРИЧ Дмитрий Олегович, КЛЮЧЕВЫЕ СЛОВА: прогнозирование; профессиональная дея-
г. Санкт-Петербург, Россия, [email protected] тельность; выпускник военно-учебного заведения; нечеткая ней-
ронная сеть; регрессионный анализ..
ОХОТНИКОВ Юрий Юрьевич,
г. Санкт-Петербург, Россия, [email protected]
ШАЙМУХАМЕТОВ Шамиль Ильдусович,
г. Санкт-Петербург, Россия, [email protected]
АННОТАЦИЯ
В современном мире большое значение приобретает умение специалиста адаптироваться к динамически изменяющимся условиям своей профессиональной деятельности. Эта задача очень актуальна при подготовке высококвалифицированных кадров в высших военных образовательных учреждениях Министерства обороны Российской Федерации. Высокая стоимость обучения квалифицированных военных специалистов, высокий уровень требований, предъявляемых к результатам их профессиональной деятельности, обусловливает чрезвычайно высокую важность решения задачи прогнозирования и раннего оценивания успешности дальнейшей профессиональной деятельности выпускников вузов Министерства обороны Российской Федерации. Успешность профессиональной деятельности выпускника определяется соответствием профессионально-
важных качеств требованиям, предъявляемым к его будущей военно-профессиональной деятельности.
Наиболее предпочтительным математическим аппаратом для моделирования подобного класса задач, где имеется очень много нечетко выраженных входных данных, в совокупностях которых скрыты закономерности и взаимосвязи между ними, является аппарат нечетких нейронных сетей. Целесообразность использования нечетких нейронных сетей также обусловлена неполной или нечетко выраженной информацией предпочтений, а также интуитивно формулируемыми правилами решения таких задач.
Для реализации процесса оценивания результатов деятельности и прогнозирования успешности выпускника предлагается к рассмотрению класс адаптивных сетей функционально экви-
валентных системам нечетких рассуждений. Подобная архитектура носит название ДЫПБ. ДЫПБ является одним из первых вариантов гибридных нейро-нечетких сетей - нейронной сети прямого распространения сигнала особого типа. Архитектура нейро-нечеткой сети изоморфна нечеткой базе знаний. В нейро-нечетких сетях используются дифференцируемые реализации треугольных норм (умножение и вероятностное ИЛИ), а также гладкие функции принадлежности. Это позволяет применять для настройки нейро-нечетких сетей быстрые алгоритмы обучения нейронных сетей, основанные на методе обратного распространения ошибки. ДЫПБ реализует систему нечеткого вывода в виде пятислойной нейронной сети прямого распространения сигнала.
Использование предлагаемого подхода поможет с выбором наиболее адекватных и исключения малоинформативных методик профессионального и психологического отбора, с селекцией наи-
более результативных методик обучения. Различные предпочтения могут обосновываться применением методов одномерной и многомерной статистики. После этого проводится разработка алгоритма (решающего правила) оценки профпригодности. Наиболее часто для этих целей используют множественный регрессионный анализ, основанный на связях психофизиологических свойств с «внешними критериями», под которыми понимаются качество (успешность) обучения или деятельности.
СВЕДЕНИЯ ОБ АВТОРАХ:
Петрич Д. О., к.т.н., старший преподаватель Военно-космической академии имени А.Ф.Можайского;
Охотников Ю. Ю., преподаватель Военно-космической академии имени А.Ф.Можайского;
Шаймухаметов Ш. И., адъюнкт Военно-космической академии имени А.Ф.Можайского.
Для цитирования: Петрич Д. О., Охотников Ю. Ю., Шаймухаметов Ш. И. Применение нечетких нейронных сетей в прогнозировании успешности профессиональной деятельности военных специалистов // Наукоемкие технологии в космических исследованиях Земли. 2017. Т. 10. № 2. С. 100-106. doi 10.24411/2409-5419-2018-10045
