You get a bonus - 1 coin for daily activity. Now you have 1 coin

The use of hybrid systems of computational intelligence to select a rational option for management decisions

Lecture



The use of hybrid systems of computational intelligence to select a rational option for management decisions

SUMMARY: The analysis of the current state decision theory, recommendations were made to improve this process using a hybrid system developed on the basis of modern information technologies.

KEYWORDS: decision making method, computational intelligence, hybrid system, intelligent system, neural network.

SYSTEM research in various fields of human activity began to develop intensively in the middle of the last century, when a synthetic scientific discipline - the study of operations - appeared on the basis of theories of efficiency, games, mass service and other theories . As a result of these studies, a theory of systems emerged, part of which is decision theory. With its help, great successes have been achieved, however, traditional methods of analysis that use the experience, intuition and the ability of management officials (DL OU) to associate, that is, which are extremely bad, are still crucial in the study and management of complex systems. can be described using formal mathematical methods.

Currently, decision theory makes extensive use of criteria-based and situational methods. Moreover, the difference between the situational method and the criterial one is that it does not establish the fact that the alternatives are not equivalent, but the fact that the current situation is equivalent to one of the situations listed in the catalog.

The relevance and widespread use of the current situational method is due to the following circumstances:

in complex systems and procedures, an adequate mathematical description of the decision-making process is either absent or cumbersome;

decision-making algorithms are developed mainly for operational application (real-time operation), therefore, the use of exact optimization methods inherent in the criteria-based approach are excluded due to their complexity;

in a number of complex practical problems, a solution is developed as a result of a series of successive iterations; therefore, when implementing them, it is necessary to take into account the qualitative information coming from experts and presented, in most cases, informally.

Creation of decision-making methods for difficultly formalized tasks (i.e., tasks that cannot be numerically defined, whose goals cannot be expressed in terms of a well-defined objective function, whose algorithmic solution either does not exist or cannot be applied due to computational limitations resources) can be provided by improving the tools of situational choice, focused on the widespread use of heuristic methods. These methods should fit well with modern decision theory, constantly changing under the influence of its tools. Currently, the totality of such heuristics, as applied to the problem of choosing a solution option, is united under the general name - artificial intelligence.

Analysis of the current state of decision theory allows the following conclusions:

The first. One cannot overestimate the value of mathematical methods and assume that formal methods of mathematics are a universal means of solving problems in the management of military, industrial, economic and other types of activities.

Second. Methods based on the results of experience and intuition will be relevant for a long time.

Third. Plausible reasoning (from the standpoint of “common sense”) helps to form mathematical models based on accumulated experience in developing various models. From a formal point of view, these mathematical models can be considered as some system of axioms. In other words, these models have a body of knowledge that determines the relationship between various observations of phenomena in accordance with the fundamental theory, but not directly following from it. Recently, this approach has become generally accepted when conducting various systemic studies in the military, economic, social and other subject areas.

Fourth. Military, economic and social systems are manageable, therefore, a management goal should be formed for them, which obviously leads to the concept of a program that is outside the model, the formation procedure of which cannot be completely formalized. In addition, heuristic elements are also present in the concept of a quality criterion (efficiency), which allows one to choose a rational solution from among the admissible ones. Consequently, heuristic procedures and methods in system research and in specific tasks are and will be of great importance.

Fifth. Heuristic procedures and methods, without which it is impossible to imagine the functioning of complex control systems, are decision-making methods that use accumulated and generalized experience.

Sixth. Informal and rigorous mathematical methods of analysis cannot be opposed, since a decision must be made on the basis of a combination of both ways of thinking.

In addition, it is necessary to take into account the fact that each VCA has the following features: unsteadiness (variability) of individual parameters of the system and stochasticity of its behavior as a whole; uniqueness and unpredictability of system behavior in specific conditions; the formation of various behaviors, due to the presence in the system of active elements - people; the ability to self-regulate and adapt to changing conditions; ability and desire for goal formation (since in systems with active elements goals are formed within the system itself). This is evidenced by the site https://intellect.icu.

For a long time, they tried to describe the accounting of the given features using classical systems of artificial intelligence, i.e. systems based on symbolic calculations and other formal methods. However, it soon became clear that with the help of symbolic information processing, in most cases, it is not possible to solve the applied problems of decision support systems (primarily military) if it is impossible for them to obtain all the necessary information.

As studies conducted in the 1980s and 90s have shown, the solution to this situation is the use of systems based on computational intelligence (in foreign literature - “Soft Computing”).

Computational intelligence (VI) is understood as a scientific direction where artificial intelligence problems are solved on the basis of new non-traditional methods of computation, and VI technology is understood as a combination of new methods and means of processing knowledge, workflow, methods for developing and choosing alternative solutions combined into an integrated technological system to make and communicate decisions to performers. This totality assumes, as a rule, the presence of a developed human-machine interface, a system (or elements) of a VI and the possibility of using electronic maps of the area.

Currently, it is believed that VI includes the following main methods:

neural networks - using training, adaptation, classification, system modeling and identification of systems based on source data;

fuzzy logic - based on the theory of fuzzy sets and providing effective means of mathematical reflection of the uncertainty and fuzziness of the source information, allowing you to build a model that is adequate to the studied subject area;

genetic - using synthesis, tuning and optimization of the studied systems using a specially organized random search and evolutionary modeling.

These methods are the main ones in the VI, however, it should be noted that the number of new methods that have joined them in recent years is constantly expanding, not being strictly defined. The most significant of them are listed below:

cognitive computer graphics - data visualization methods that enable the activation of visual-figurative thinking mechanisms of decision makers, facilitating decision-making in a difficult environment or finding a solution to a complex problem;

fractal geometry; chaos theory; nonlinear dynamics.

To date, there is a fairly large number of diverse classifications of modern information technologies that take into account the paradigms of the subject areas under consideration. With regard to the field of military control, one can propose the classification shown in the figure.

The main characteristics of intelligent systems (IP) based on VI, determining their application in the field of military command, are as follows:

the ability to learn and self-learn, that is, the ability of IP-based ICs after presenting the input information to self-adjust, providing output information with the required accuracy, in contrast to systems based on symbolic calculations not capable of self-learning in principle;

the ability to adapt, i.e., quickly change their parameters in accordance with a changing environment;

“Transparency” of the explanation, that is, to represent the knowledge extracted from the data in an understandable format for the expert or decision maker;

the ability to discover new things, that is, to reveal previously unknown, hidden connections and relationships in large arrays of numerical, textual and visual information, to predict the emergence of new processes and trends;

The use of hybrid systems of computational intelligence to select a rational option for management decisions

Classification of modern information technology

nonlinearity, i.e., to approximate arbitrarily complex nonlinear functions with any predetermined accuracy;

universality, i.e., to solve a wide range of problems and be free from any assumptions regarding the data source;

parallelism, i.e., the ability to parallelly process information;

stability, that is, the ability to continue to complete a task while maintaining a given quality of decisions, in conditions when part of its structure is damaged;

creativity, i.e., the ability to generate new (not encountered during training) options for solving the problem.

The listed characteristics are in varying degrees inherent in each of the ICs using the specific VI method, and therefore, they all have strengths and weaknesses that make it possible to create hybrid intelligent systems from them, combining the positive properties of both individual ICs based on VI and existing classical methods . At the same time, the components of such a hybrid system do not compete with each other but create a synergistic effect of mutually reinforcing their strengths and weakening weaknesses. So, for example, the representation of knowledge in neural networks (NS) in the form of weight matrices does not allow the analysis of the results obtained, while in output systems based on fuzzy rules, the results are interpreted as reverse output protocols. NSs are trained using a universal algorithm in which the laborious extraction of knowledge is replaced by the preparation of a sufficient training sample in volume. For fuzzy inference systems, the extraction of knowledge includes complex processes of formalizing concepts, defining membership functions and creating inference rules. At the same time, fuzzy NSs are trained as ordinary NSs, and their results are explained as in fuzzy inference systems.

According to the results of studies conducted in Russia and abroad, the core of a hybrid IS based on VI is most likely to be NS, which is ahead of other VI systems in such parameters as the ability to learn, generalize and adapt. Based on this, to solve the problems of classification and clustering, multidimensional optimization, visualization, lowering dimensionality, making decisions based on incomplete and inaccurate data, etc., if necessary, you can use expert and fuzzy systems integrated with NS, genetic algorithms, cognitive computer graphics, fractals, chaos theory, nonlinear dynamics, etc. In addition, it is possible to use such a hybrid IS based on VI to replace complex calculation methods (not working both in real time and with incomplete and inaccurate data), by training it on existing results; building models of various subject areas, based on the accumulated arrays of experimental data, without their analytical description; solutions to various difficult and unformalized tasks.

It should be noted that the hybridization of IP currently refers to the integration of methods and technologies at a level that involves the creation of new methods using the concepts of combined basic methods. At the same time, the components of a hybrid IP should provide such properties as uncertainty management, learning ability and self-adaptation.

An example of the construction and use of the simplest hybrid IC based on the Hamming neural network and the anamorphic algorithm (a transformation that transforms a visual image constructed in the Euclidean metric into a visual image built on the basis of the selected anamorphic metric), which can be applied to cognitive computer graphics methods, can be considered using an example creating a system of engineering barriers (PPE) in a defensive operation (GS).

It is well known that the decision to create PPE in the PA consists not only of an extremely complex and time-consuming stage of preliminary formation of solution options (initial set of alternatives) by experts and / or decision makers, but also of a no less time-consuming and responsible stage - the stage of choosing the most similar solution from the initial set of alternatives, taking into account the task, the current situation, the available forces and means. This requires a constant search for new methods and technologies that allow, in the general case, a contradiction between the increase in the information complexity of the preparation and choice of a solution option and the time allotted for this.

The choice of the solution for the creation of PPE in the PA is always associated with the need to obtain and analyze extensive and specific information. Basically, this is information about the characteristics of the terrain within the combat zone and the objects located on it, the results of assessing patency, protective properties of the terrain, analysis of the road network, etc. Moreover, at the level of the Chief of Engineering Troops (NIV), generalized, averaged ( taking into account the possible scatter) information that ensures the speed of decision-making while maintaining its specified reliability.

Topographic maps carrying limited information, various electronic terrain maps (ECM) are used as one of the source of initial data in the problem under consideration, and additional information about individual terrain sections is obtained during air or ground reconnaissance.

The principle of operation of the proposed hybrid system is as follows.

The first one. The neural network selects from the library of pre-created visual images (drawn on the electronic map of the area (ECM) of the graphic parts of the decision, with an explanatory note calculated for each option), the decision to create a PPE at 00 is one solution that is most appropriate for the actual situation. The choice of NS Hamming is determined by the fact that this multilayer network of associative memory with negative fixed connections between the individual layers is characterized by a smaller number of calculations and the amount of computer RAM used in the classification of visual images. The meaning of her work is to find a library visual image with a minimum Hamming distance (the number of differing values ​​in two binary vectors) to the input version of the decision to create a PPE in accordance with the current situation, with the activation of only one network output that is most suitable for this image.

The second one. For the stable operation of the NS, library visual images of the decision to create PPE at 00 and the input visual image reflecting the current situation are converted using the anamorphic algorithm, and the delay time on the PPE elements is taken as the anamorphic index. The instability of the NS, in general, can be caused by the fact that different library solutions, consisting of areal figures - PPE elements, can have the same or close arrangement of component elements with a different delay time for each of them. The National Assembly will not be able to classify the input visual image if the similarity measure between it and the library visual images is the same or small enough, therefore it will be forced to make a decision based on a random choice between them. As a result of anamorphing, the geometric shape of all elements of all visual images will change, and their anamorphic indices become equal. This transformation will ensure that all library visual images are distinguished from each other and, as a result, ensure stable NS operation, the ability to analyze the properties of PPE elements in the space of a selected anamorphic indicator (in particular, delays on PPE elements), develop scenarios for the development of military operations and correspond to each of them the evolution of PPE.

As a result of submitting to the input of the described hybrid system a solution option developed by the DL OU and corresponding to the actual situation, one of the library solution options will be presented at the system output, which is as similar as possible to the input solution option and the corresponding explanatory note with PPE calculations. Also, the ECM will show the differences between the library solution selected by the neural network for creating PPE at 00 from the input developed by the DL OA.

After evaluating the obtained version of the decision and clarifying it, it is submitted as proposals to the official (NIV) making the decision on the creation of personal protective equipment. If the situation and the time budget allow, then the refinement option can be repeatedly applied to the input of the hybrid system until a rational solution is obtained for creating PPE at 00.

The proposed hybrid system makes it possible to reduce by an order of magnitude the decision-making time of the DL OU, use the accumulated experience and informal knowledge, take into account the intuition of experts when compiling a library of solution options. However, it is necessary to take into account the fact that the work of compiling such a library is a rather difficult task, requiring additional research and should be carried out in the course of planning the defensive operation in advance.

Zakharov I.G. Justification of the choice. Theory of practice. - SPb .: Shipbuilding, 2006. 232 S.

Razumov O.S., Blagodatskikh V.A. Systemic knowledge: concept, methodology, practice. - M.: Finance and Statistics, 2006. 400 S.

Frolov Yu. V. Intelligent systems and managerial decisions. -M .: LNA, 2000. 294 S. Balyberdin B. A., Penkin OM, Polunin A, I. Problematic issues of the creation and implementation of new information technologies in automated military systems. -M.: Armament. Politics. Conversion, 2001.146 S.

Tarkov M. S. Neuro-computer systems: A manual - M.: Internet University of Information Technologies; BINOMIAL. Knowledge laboratory. 2006.140 S.

Galushkin A.I., Ivanov V.V., et al. Some Conceptual Issues of the Development of Neurocomputers - Foreign Radio Electronics, No. 2, 1997. P. 3 - 10.

Khaikin S. Neural networks: a full course. 2nd ed. Per. from English M.: Publishing House "William", 2006. 1104 S.


Comments


To leave a comment
If you have any suggestion, idea, thanks or comment, feel free to write. We really value feedback and are glad to hear your opinion.
To reply

Computational Intelligence

Terms: Computational Intelligence