Intelligent system of identification of local area network state

. This article considers the problem of diagnosing local area network, as well as the task of structural determination is the creation and easy adjustment of the model object by building the knowledge base of available expert information. The algorithm of formalization of knowledge in designing intelligent systems is reviewed and proposed. The article considers the theory of fuzzy sets and fuzzy logic. The concept of the theory of fuzzy sets and fuzzy logic are formalized in the form of fuzzy and linguistic variables, and the vagueness of certain operations in the overall decision-making process is synthesized in the form of fuzzy algorithms. For the effective solution of tasks of type analysis and hypothetical sources of network problems in a local area network, a method of constructing membership functions based on analytical processing of the results of expert surveys is used.


Introduction
When designing an intelligent system, the problem of acquiring and presenting expert knowledge always arises. This specific design feature is associated with the complexity of the process of selecting analytically important facts and rules, the subsequent structuring and construction of a knowledge base from them.
Nowadays, to formalize this knowledge, the apparatus of the theory of fuzzy sets (TFS) and fuzzy logic (FL) is being successfully used. The concepts of TFS and FL are formalized in the form of fuzzy and linguistic variables, and the fuzziness of individual operations in the general decision-making process is synthesized in the form of fuzzy algorithms. An algorithm for constructing an analytical model of a fuzzy intelligent system for identifying the states of a local area network (LAN) is shown in Figure 1 [1].

Materials and method
Let's think of a LAN as the OSI seven-layer reference model. This view simplifies the task of moving information between computers across a LAN environment and presents seven smaller analytical subtasks. Each of these subtasks is solved using the corresponding level of the reference model of the local area network (physical, channel, network, transport, session, representative and application levels). Therefore, the main task of diagnosing a LAN 0 Z can be represented by a certain composition of seven independent and autonomous tasks 7  X .
An important condition for the guaranteed solution of the presented problem is the existence of a functional dependence To determine this dependence, we will define the input variables , as well as the output variable y as linguistic variables (LV) determined using universal sets [8]: and y we will use terms from the following term sets: Possible difference in the cardinality of term sets We will identify the terms of input and output LVs as fuzzy sets (FS) that can be specified on a universal set i \ , 8 . It will be determined by relations (7) -(10).
and y are numerical variables, then fuzzy sets and y are considered as qualitative variables, fuzzy sets and Y are expressed by formulas (9) and (10).
This approach for building a fuzzy model is called fuzzification of linguistic variables.

Results
In accordance with (6), let us choose the MISO-structure ("Multi Input -Single Output") [5] of the fuzzy knowledge base. It is also necessary to select N experimental data to link the outputs and inputs of the studied object. Let's distribute them according to the following principle: To take into account the different degrees of the expert universality when displaying the rules, we will use weighting factors. When entering weighting factors of the rules into a fuzzy knowledge base the following transformations will occur:

Discussion
Using the knowledge matrix given in Table 1 or an equivalent system of logical statements (17), we define a system of fuzzy logical equations that determine the values of the membership functions of various solutions for fixed values of the input variables: where -logical AND, -OR.  A fuzzy set g for defining a linguistic term j g on a universal set V can be represented as: It is necessary to find the degree of membership of each element of the set V to the elements of the set G. In other words, it is necessary to find ) Two existing methods can be distinguished, with the help of which it is possible to construct membership functions. The first method is to statistically process the opinions of a group of experts. In the second method, pairwise comparisons are made, which are performed by one expert [3].
Effectively solving the problem of analyzing the type and hypothetical sources of LAN To get more accurate estimates, sometimes reviewed experts are divided into categories of experience, with assigning each of them their own weight.
The following membership functions are widely used in practice: triangular, Gaussian (bell-shaped), trapezoidal and sigmoidal.
Membership functions are often specified in parametric form. For this, the parameters of the membership function are determined. These parameters include the core, level, maximum coordinate and concentration coefficient.

Conclusion
The idea of an algorithm for solving this problem is to apply L. Zadeh's "compositional rule of inference" [1,5], which established a connection between variables -one input and one output. "This rule is generalized to the system of one output and n inputs" [1,5], which corresponds to the full knowledge matrix (table 1).
The scheme of the fuzzy inference process includes three stages ( Figure 2) [1]: the introduction of fuzziness (fuzzification), fuzzy inference and reduction to clarity (defuzzification). To optimize a fuzzy knowledge base, it is necessary to carry out two stages (as in the case of the optimization of nonlinear objects) -the stages of structural and parametric definitions ( Figure 3).  At the stage of structural definition, the creation and simple configuration of the object model is carried out. This is possible using the method of building a knowledge base using the data obtained as a result of collecting expert information. To simplify the adjustment of weighting factors and forms of membership functions, methods of paired comparisons are used (for example, Saaty, Cogger) [7]. The higher the professional qualities of an expert and the more accurately he makes a forecast, the higher the adequacy of the fuzzy model built at the stage of simple setup.
At the same time, it is impossible to speak unambiguously about the guaranteed coincidence of the results obtained by the theoretical method, using fuzzy logic and data obtained experimentally. Thus, the algorithm should provide for the second stage, at which the fine tuning of the fuzzy model is carried out by training it using experimental data.
This stage (fine tuning stage or parametric determination) is formulated as a nonlinear optimization problem. The problem posed can be solved by various nonlinear programming methods (for example, gradient methods, quadratic programming method) with the use of neural LANs, genetic algorithms, etc.