INFORMATION AND WEB TECHNOLOGIES

. Neural networks are well suited for pattern recognition and classification. Matlab provides an extensive library and toolbox in neural networks. In the study, I designed a neural network using the Matlab Neural network toolbox. I carried out this process by changing various settings to get better performance and result. Recently, the processing and analysis of large volumes of data has become an urgent task. "Big Data" can be defined as "a large amount of unstructured data". This concept is the object of many scientific studies and developments in the field of information technology. The main characteristics of the concept "Big Data" are their unstructured nature


No 151
Introduction. In recent years, these issues are usually considered in Data Mining and Knowledge Discovery in Databases sections. More generally, such problems belong to the section of data analysis. In a broad sense, the search for dependencies based on the input and output data presented in the dataset can be attributed to the section of mathematical modeling, or rather, the identification of mathematical models.
Feedforward neural networks. In building such models, the problem of structural identification is partially solved. The development of feed-forward neural networks is an approach based on neurostructural models. This makes it possible to apply most of the methods of operation and identification of neural networks to the NSM class without significant changes. In this case, the structure of a certain model class from NSM can be effectively considered.
An NSM is a collection of interconnected neuron-like elements (NPE). NPE is a generalization of the artificial neuron concept. Like a neuron, an NPE generally transforms a vector input into a scalar output. In the neuron, first, a weighted summation of the input signals is performed, and then some, usually non-linear, activation function is applied. NPE, unlike a normal neuron, can perform an arbitrary dependence.
NSMs include not only feed-forward neural networks, but also neural networks with non-classical activation functions, probabilistic neural networks, neural networks with radial basis functions, fuzzy Takagi-Sudgeno models with differentiable inference operations, and neuro-there are fuzzy models. The purpose of this work is to build identification algorithms for a special class of neuro-fuzzy models of the ANFIS structure, taking into account the characteristics of the learning problem [6].
ANFIS structural models are one way to combine fuzzy logic and NSPR. It belongs to the class of hybrid neuro-fuzzy systems (Neuro-Fuzzy Systems, NFS). Hybrid models allow combining the advantages of fuzzy inference systems with the advantages of NSPR. A feature of ANFIS neuro-fuzzy systems is their tractability These models can also be easily interpreted as fuzzy No 151 knowledge representation models. Let us consider the method of designing neuro-fuzzy ANFIS architectures using the example of systems functionally equivalent to Takagi-Sudgeno systems. ANFIS structure has the following general form [5]: If xj*R, j=1,…,n, -output system; yi*R, i=1,…,m,individual rule outputs; Aij are parameter-dependent membership functions {aij}; fi -functions depending on system inputs and parameters {bi}. ANFIS structural models are a combination of linguistic and analytical models. Input values and output are real values. Often, analytical functions resulting from rules have a linear structure in parameters [4]: Let's continue the study with an example. Effective organization of transportation of raw materials for the production of textile products in conditions of environmental dynamics and uncertainty is related to high-level management of logistics. In this regard, depending on the requirements for the delivery process, the problem of creating a new textile transportation organization that ensures the stability of the activity of individual parts of the delivery logistics chain arises.
The purpose of the study is to increase the efficiency of the logistics organization of transportation for textile companies. Achieving the goals is possible by using methods of fuzzy set theory and artificial neural networks that allow us to take into account various inaccuracies and uncertainties arising in real-time transportation management processes.
It is proposed to apply a fuzzy inference system in the No 151 form of a neuro-fuzzy network to effectively coordinate the interaction of different types of transport at certain stages of the transport process of organizing real-time transportation in the textile industry.
A neural network is a directed graph where external input or output variables and arcs are signal propagation or synapses. The number of neuron layers depends on the specific formulation of the problem, and the number of connections between neurons is not limited [3].
The process of organizing transportation in the textile industry is considered the transportation of raw materials for the production of textile products to the destination by the choice of transport mode and route. Transport organization is characterized by a set of indicators that reflect the level of transport service, which are fuzzy linguistic variables at the project development stage, so it is appropriate to use the fuzzy sets apparatus [2].
Fuzzy linguistic variables are used as input during the development of a neuro-fuzzy network: type of transport, speed of transportation, the safety of cargo, dependence on climatic conditions, carrying capacity, and cost of cargo delivery [1].
Let's determine the term sets of the variables with the following verbal-numerical scale (Table 1): It is required to choose the most rational route with satisfactory transport service at the required level.
An approach based on ANFIS fuzzy neural network in MATLAB software is proposed to solve the problem [8].
ANFIS is an abbreviation of AdaptiveNeuro-FuzzyInferenceSystem -(Adaptive Neuro-Fuzzy System). ANFISeditor allows to automatic synthesize of neuro-fuzzy networks from experimental data. A neuro-fuzzy network can be considered one of the types of Sugeno-type fuzzy inference systems. At the same time, the membership functions of the synthesized systems are adjusted (trained) in such a way as to minimize the deviations between the results of fuzzy modeling and the experimental data [9].
The general sequence of the hybrid network model development process can be presented in the following sequence.
1. Preparation of a file with training data. It is advisable to use MS Excel, a spreadsheet editor. The training set should be saved in an external file with the *.dat extension.
2. The ANFIS editor opens and the file is loaded with the training data. LoadData command When the data load button is clicked, if the data is loaded from disk, a file selection dialog box appears. The graphical interface of the ANFIS editor with loaded training data is depicted in Fig. 1 [7].
After preparing and loading the training data, the structure of the FIS Sugeno-type fuzzy inference system with a hybrid network model is created in the Matlab system. For this purpose, use the Generate FIS button located at the bottom of the editor's working window. When choosing the grid partition method, a window appears for entering the parameters of the lattice method, where we must specify the number of  and the type of membership functions. After creating the hybrid grid structure, the Structure located on the right side of the graphics window can visualize the structure by pressing the button [6].

Figure 1
Loaded training data of the graphical interface of the ANFIS editor Before training the hybrid network, we need to set the training parameters, for this we need to use the following group of options in the lower right part of the working window [5]: 1) choose a hybrid network training method -hybrid (hybrid), which is a combination of the least squares method and the inverse gradient reduction method.
2) set the learning error level (ErrorTolerance)default value is 0 (not recommended to change).
3) the number of training periods is determined. The progress of the training process is depicted in the visualization window in the form of a graph of the dependence of the error on the number of training cycles.
Further tuning of the parameters of the built and trained hybrid network can be done using the standard graphical tools of the FuzzyLogicToolbox package. For this purpose, it is recommended to save the created fuzzy inference system in an

Orléans, France
No 151 external file with the *.fis extension, after which this file should be loaded into the FIS fuzzy inference system editor. Using the graphical interface FIS Editor, we change the name of the linguistic variables and change the names of the terms of the linguistic variables [4].
The next stage is the testing of the fuzzy system with the output of the results to the field of visualization. Data for the test is loaded (Loaddata). It is done by selecting the test. In the Test field (TestFis), the Testingdata option is selected and TestNow starts testing.
6. To check the adequacy of the established fuzzy hybrid network model, it is necessary to use the function evals. Obtaining the prediction result for certain input data is done by entering the following code in the command field: x=[1 7073703000]% Enter input parameters fis = reads('set1.fis');% Loading generated fuzzy inference system file y = evals(x, fis)% Prediction output For the membership function y=0.7, the result value of the output linguistic variable "satisfaction of transport service" is the result of solving the fuzzy inference problem for the proposed values of the input linguistic variables of the transport organization indicators [2].

Conclusion
The obtained results give reason to talk about the possibilities of the practical application of the obtained neuro-fuzzy network for increasing the efficiency of the organization of textile transportation. Further research involves the development of fuzzy situational networks for operational decision-making while managing the interaction of all participants in a real-time transportation system.
1. The proposed approach of the coordinated interaction of the elements and connections of the raw material transportation system for the production of textile products allows for reducing the transportation time itself.
2. The use of the ANFIS editor in the MATLAB environment is very promising for solving the problem of coordinating the links of the logistics chain and organizing the transportation of raw materials for the production of textile products. The disadvantage is that the quality of the results depends on the quality of the experimental data or training samples.

No 151
Therefore, the selection of training samples is an important process when using ANFIS.
A neuro-fuzzy model whose structure is determined by the following parameters was developed to form the rule system of the description form of this developed neuro-fuzzy model: 1) the number of input parameters object (determines the number of neurons in the input layer of the network); 2) the number of values (gradations) of the input parameters of the object (determines the number of neurons in the valuegradation layer of the input neurons of the network); 3) the number of values of the target parameter of the object (determines the number of neurons in the value layer of the output neuron of the network); 4) a result algorithm based on a system of rules of the form (determines the number of network layers and their functionality).