Використання когнітивного моделювання при управлінні біотехнологічними об’єктами харчових виробництв

Анотація

USE OF COGNITIVE SIMULATION AT THE MANAGEMENT OF BIOTECHNOLOGICAL OBJECTS OF FOOD PRODUCTION

N. Zaiets, V. Shtepa

Currently, the prediction of the state and some parameters of the biotechnological objects of food production is associated with the involvement of a group of experts in this subject area. A convenient tool for aggregating the knowledge of a large and, as a rule, highly distributed expert group is the mathematical apparatus of cognitive maps and fuzzy cognitive maps (NCCs). The cognitive map is represented as a signed digraph. The advantages of using NCC, compared with conventional cognitive maps, are visibility, which is an important factor in the work of the expert group of the subject area and the ability to numerically describe processes that are modeled. Since most of the actual information on the control object is derived from expert assessments, it is largely subjective [1]. Moreover, the opinions of experts on the same issue can differ substantially, sometimes fundamentally. Therefore, the problem of optimal generalization of expert opinions with a view to constructing an adequate NCC is relevant.

The methodology of cognitive modeling, designed for analysis and decision-making in unforeseen situations, is based on modeling subjective representations of experts about the situation and includes: methodology for structuring the situation;model for representing expert knowledge in the form of a signed digraph (cognitive map) (F, W), where F is the set of factors of the situation, W is the set of causal relationships between the factors of the situation; methods for analyzing the situation.

The methodology of cognitive modeling is developing in the direction of improving the apparatus of analysis and modeling the situation. However, the existing methodology for structuring the situation and the model for representing the expert's knowledge does not allow us to analyze complex situations. Creation of large models, including tens or hundreds of factors, requires the development of another model for presenting knowledge about the situation, the methodology for structuring unforeseen complex situations, the methods for explaining and interpreting modeling results and supporting the generation of solutions.

One of the effective means of solving this situation is fuzzy cognitive maps, where a lot of connections between concepts are presented in the form of numerical values of the degrees of causality of such links. On the basis of the constructed NCC, a matrix of mutual influences of concepts is formed, after which the behavior and reliability of the map are examined. Its system characteristics - consensuses, dissonances - are calculated.

For the use of cognitive maps, as an instrument for aggregating the knowledge of an expert group, it is necessary to establish the exact values of fuzzy variable relationships between factors, and it presents difficulties in creating a cognitive map with a large number of vertices. To automatically adjust the values of variable relationships, cognitive map learning algorithms are used. The task of teaching the cognitive map is to minimize the error of the prediction result. The prediction error is estimated as an error between the predicted indicators of the map and the real values of the factors known from historical observations.

The aim of the study is to develop a fuzzy cognitive map structure, which includes a fuzzy neural network of expert evaluation synthesis with the purpose of adaptive adjustment of the matrix of mutual influence of NCC concepts.

In further studies, we use generalized NCCs, each concept of which is characterized by the term-set of the linguistic variable.

The block diagram of the algorithm of creation and practical use of the fuzzy system of generalization of expert estimation in the regular mode is constructed. However, one of the main weaknesses of systems based on fuzzy logic is their inability to self-learn, for their adjustment it is necessary to recruit experts at full functional stop. To solve our problem we need the ability to self-adapt when changing expert estimates or parameters of an object.

Therefore, it makes sense to use the apparatus of fuzzy neural networks - neural networks with clear signals, weights and activation functions, but combining them using t-norm, t-konorma or other operations.

The software module for calculation of the weight coefficients of the NCC can be created and investigated in MatLAB system, integrated into the corresponding mathematical apparatus of fuzzy neural networks. The structure of the fuzzy cognitive map was developed on the basis of experimental studies and object-oriented analysis of the brewery.

The task of the NCC is to identify effective strategies and control scenarios for a given technological complex. The formation of values of weight coefficients based on expert estimates should solve the problem of the impossibility of an expert survey of experts in changing the parameters of the complex operation.

The corresponding structure of the fuzzy cognitive map was developed in cooperation with three experts in the brewing industry. Created by the NCC, which will function in accordance with the simplified algorithm of the fuzzy output, will allow scenario to investigate the behavior of the system when changing the values of the concepts.

According to the developed methodology and expert evaluations of the three specialists, we form the values of the coefficients of the NCC matrix. The maximum difference between the thoughts of qualified experts was 30%. It is also clear that a simple reduction to the arithmetic mean value will not be correct, as it will lead to the actual loss of the essence of expert judgment.

The article substantiates and develops the method of adaptive formation of the matrix of mutual influence of the concepts of the fuzzy cognitive map of the brewery on the basis of the use of a fuzzy neural network. Based on the technological features of the investigated object, the interval is set through which scenario planning is performed. The software being developed includes the implementation of the work of both modules, provides the condition of adaptability: when the system exits beyond the established efficiency, it is possible to retrain the system.