Neuro-fuzzy Technologies Research Articles

An example of intelligent greenhousing synthesis based on modern digital systems, neuro-fuzzy technologies and IoT

Challenges in development of new generation modern control systems based on discrete logic are reviewed in this article. System analysis of the current characteristic and properties of data processing systems is given: the problem of conversion accuracy and categorization of new criteria of new system effectiveness. The research experience of reviewed subject is summarized, a processing algorithm for greenhousing hardware control system with support for decision-making on monitoring the operation of sensors, taking into account the neuro-fuzzy technologies and graph theory, in the context of the algorithm theory and solving the problem of calculating the critical path for the quantifier of interaction with the subsequent processing in a neuro-fuzzy application in a quantitative ratio is given. There is description of generalizable algorithm for the constructed software control model with six sensors and a general assessment of the effectiveness of the traditional way for formal model of the greenhouse automated control system and the proposed way based on the balance model in the article.

Model of Assessing the Overdue Debts in a Commercial Bank Using Neuro-Fuzzy Technologies

This article considers the problems of overdue credit debt and the creation of effective methods to manage problem debts in banks. The purpose of this paper is to study the problem of overdue credit debt and create effective methods to manage problem debts in financial institutions. Based on a combination of tools of fuzzy logic theory and artificial neural networks, an economic-mathematical model of collection scoring was built. Kohonen self-organizing maps were used to set the parameters of membership functions in the process of fuzzification of quantitative variables of the built model. Data were taken from the official websites of four Bulgarian banks for 2015–2019. The volume of the prepared sample amounted to 1000 credit agreements with active overdue payments. The practical value of the built model of collection scoring for the recovery of overdue debt lies in the possibility to make recommendations for work with each segment of the portfolio of overdue loans in accordance with the calculated level of credit risk. The introduction of credit risk assessment models based on neuro-fuzzy technologies in the work of financial institutions will have a positive impact on the financial results of lending activities of banks.

Control of Organic Rankine Cycle, a neuro-fuzzy approach

Undesirable emissions in internal combustion engines are of major concern due to their negative effect on the human health and global warming. Implementing an organic Rankine cycle (ORC) in mobile applications can help to mitigate their negative effect by transferring some of their wasted heat into useful electrical or mechanical energy. Considering the stringent regulations in automobile industry ensuring the safety of equipment before implementing them on the vehicles is vital. In this study, a novel control approach is proposed to ensure the safe operation of Organic Rankine Cycle (ORC) waste heat recovery (WHR) system and stabilize its work output when subjected to transient heat sources in a range of waste heat from heavy-duty diesel engines. The control strategy comprises a neuro-fuzzy controller based on the inverse dynamics of the ORC system to control the superheating at the evaporator outlet by adjusting the pump speed and a PI controller to maintain the expander work output by regulating the mass flow rate at the expander inlet. The performance of the control strategy is investigated with respect to set-point tracking and its robustness is tested in the presence of noises. The simulation results indicate an enhancement in the controller performance by combination of feedforward and feedback controllers based on neuro-fuzzy technologies. The proposed control scheme not only can obtain satisfactory transient response under various loading conditions, but also can achieve desirable disturbance rejection performance.

APPLICATION OF THE THEORY OF OPTIMAL SET PARTITIONING BEFORE BUILDING MULTIPLICATIVELY WEIGHTED VORONOI DIAGRAM WITH FUZZY PARAMETERS

An algorithm for constructing a multiplicatively weighted Voronoi diagram involving fuzzy parameters with the optimal location of a finite number of generator points in a limited set of n-dimensional Euclidean space 𝐸𝑛 has been suggested in the paper. The algorithm has been developed based on the synthesis of methods of solving the problems of optimal set partitioning theory involving neurofuzzy technologies modifications of N.Z. Shor 𝑟 -algorithm for solving nonsmooth optimization problems.

Construction of a multiplicatively weighted diagram of a crow with fuzzy parameters

An algorithm for constructing a multiplicatively weighted Voronoi diagram in the presence of fuzzy parameters with optimal location of a finite number of generator points in a bounded set of n-dimensional Euclidean space En is proposed in the paper. The algorithm is based on the formulation of a continuous set partitioning problem from En into non-intersecting subsets with a partitioning quality criterion providing the corresponding form of Voronoi diagram. Algorithms for constructing the classical Voronoi diagram and its various generalizations, which are based on the usage of the methods of the optimal set partitioning theory, have several advantages over the other used methods: they are out of thedependence of En space dimensions, which containing a partitioned bounded set into subsets, independent of the geometry of the partitioned sets, the algorithm’s complexity is not growing under increasing of number of generator points, it can be used for constructing the Voronoi diagram with optimal location of the points and others. The ability of easily construction not only already known Voronoi diagrams but also the new ones is the result of this general-purpose approach. The proposed in the paper algorithm for constructing a multiplicatively weighted Voronoi diagram in the presence of fuzzy parameters with optimal location of a finite number of generator points in a bounded set of n-dimensional Euclidean space En is developed using a synthesis of methods for solving optimal set partitioning problems, neurofuzzy technologies and modifications of the Shor’s r-algorithm for solving non-smooth optimization problems.

NEURO-FUZZY MULTICRITERIA ASSESSMENT MODEL

Context. The research of the actual problem of development of models and methods of multicriteria evaluation using neurofuzzy technologies is carried out. The purpose of this work is to develop a model for obtaining an aggregate evaluation of the significance of the object of study, which on the one hand uses different characteristics of the object, evaluated by quantitative indicators and on the basis of different models of representation of knowledge about the object, and on the other uses experience, knowledge and the expertise of experts in the relevant subject area. Objective. The object of the study is the process of modeling the experience, knowledge and competence of experts to quantify the object of study on the basis of neuro-fuzzy networks. The subject of the study is a neuro-fuzzy model of quantifying an object of study for decision making in expert data. Method. For the first time, a five-layer neuro-fuzzy model has been developed to derive quantitative and linguistic assessments of the object of the study using the expertise, expertise and expertise of the subject area. For the first time, it is proposed to use quan- titative estimates of the object of study (aggregated estimates using multicriteria models) and linguistic expert reasoning on a neurofuzzy network. For the first time, a model has been tested and verified for an example of assessing the risk of financing a startup project in the business expansion phase, and is also offered as a training for the neuro-fuzzy synaptic weight interval network. Comparison of the results of the study on different approaches to determining synaptic weights and real data with error detection. Results. The result of the study is a neural-fuzzy model for evaluating an object by many criteria. The developed model allows to combine quantitative characteristics of an object with expert opinions in the form of qualitative estimates. The rationality of the evaluation proves the advantages of the developed models. Conclusions. Sharing the apparatus of fuzzy sets and neural networks theory is a convenient simulation tool for multicriteria selection problems. As a rule, important information for management decision support systems comes from two sources: 1) obtaining object estimates by certain quantitative indicators, which creates inaccuracy; 2) from expert people who describe their subject matter knowledge, which creates subjectivity and uncertainty. Therefore, maintaining expert judgment and inaccurate data requires the ability to work with them. The paper deals with the scientific and applied problem of developing a model for obtaining an aggregate estimation of an object based on a neural-fuzzy network and can be applied in solving management decision-making problems in socio-economic systems.

���’������ ����ײ ������������ �������� ������ �� ��ײ����� ����������� � ����������

An approach is proposed to solve one continuous linear single-product problem of optimal partitioning of set Ω from n-dimensional Euclidean space Ån into subsets Ω1,...,ΩN by nding the coordinates of the centers τ1,...,τN of these subsets with fuzzy parameters in constraints. Real situations, which are described by models of optimal partitioning of sets, are most often characterized by a certain degree of uncertainty. In these cases, the quality of the decisions made in the optimization partitioning set problems is directly dependent on the completeness of all the uncertain factors that are signicant for the consequences of the decisions made. The class of the optimal partitioning set problem which makes it possible to consider uncertainty factors of not probabilistic-statistical nature are the optimal partitioning set problems in which either certain parameters in the model description are fuzzy, inaccurate, underdetermined or there exists an invalid mathematical description of some dependencies in the models (for example, demand functions and the cost of transporting a unit of production in innite dimensional transport problems) or the criteria themselves and (or) constraint systems are formulated inaccurately, or the optimization model admits the inuence of external uncontrolled disturbances of dierent forms and etc. The problem-solving algorithm was developed using a synthesis of methods for solving problems of the theory of optimal partitioning of sets with neuro-fuzzy technologies and modications of N.Z.Shor's r-algorithm for solving non-smooth optimization problems. The results are presented for the model problem of optimal partitioning of a set into three subsets with fuzzy parameters on constraints obtained by the developed approach. The results of the solution of the problem with the clear parameters are compared with the results when some parameters of the solved problem are fuzzy, unclear or their mathematical description is incorrect.

Algorithm for solving one problem of optimal partition with fuzzy parameters in the target functional

The mathematical theory of optimal set partitioning (OSP) of the n-dimensional Eu-clidean space, which has been formed for todays, is the field of the modern theory of opti-mization, namely, the new section of non-classical infinite-dimensional mathematical pro-gramming. The theory is built based on a single, theoretically defined approach that sum up initial infinitedimensional optimization problems in a certain way (with the function of Lagrange) to nonsmooth, usually, finite-dimensional optimization problems, where lat-est numerical nondifferentiated optimization methods may be used - various variants r-algorithm of N.Shor, that was developed in V. Glushkov Institute of Cybernetics of the Na-tional Academy of Sciences of Ukraine. For now, the number of directions have been formed in the theory of continuous tasks of OSP, which are defined with different types of mathematical statements of partitioning problems, as well as various spheres of its application. For example, linear and nonlinear, single-product and multiproduct, deterministic and stochastic, in the conditions of com-plete and incomplete information about the initial data, static and dynamic tasks of the OSP without limitations and with limitations, both with the given position of the centers of subsets, and with definition the optimal variant of their location. Optimal set partitioning problems in uncertainty are the least developed for today is the direction of this theory, in particular, tasks where a number of parameters are fuzzy, inaccurate, or there are insuffi-cient mathematical description of some dependencies in the model. Such models refer to the fuzzy OSP problems, and special solutions and methods are needed to solve them. In this paper, we propose an algorithm for solving a continuous linear single-product problem of optimal set partitioning of n-dimensional Euclidean spaces Еn into a subset with searching of coordinates of the centers of these subsets with restrictions in the form of equalities and inequalities where target function has fuzzy parameters. The algorithm is built based on the application of neuro-fuzzy technologies and N.Shor r-algorithm

Valuation of Startups Investment Attractiveness Based on Neuro-Fuzzy Technologies

Valuation of Startups Investment Attractiveness Based on Neuro-Fuzzy Technologies

Multi-topic text document classification based on the neuro-fuzzy technologies

This article discusses a problem of an intelligent text processing. Architecture of the neuro-fuzzy system is presented for classification of text documents and on-line learning algorithm for fuzzy network adaptive vector quantization.

Elastodynamic shape modeler: a tool for defining the deformation behavior of virtual tissues.

A main goal of surgical simulators is the creation of virtual training environments for prospective surgeons. Thus, students can rehearse the various steps of surgical procedures on a computer system without any risk to the patient. One main condition for realistic training is the simulated interaction with virtual medical devices, such as endoscopic instruments. In particular, the virtual deformation and transection of tissues are important. For this application, a neuro-fuzzy model has been developed, which allows the description of the visual and haptic deformation behavior of the simulated tissue by means of expert knowledge in the form of medical terms. Pathologic conditions affecting the visual and haptic tissue response can be easily changed by a medical specialist without mathematical knowledge. By using the personal computer-based program Elastodynamic Shape Modeler, these conditions can be adjusted via a graphical user interface. With a force feedback device, which is similar to a real laparoscopic instrument, virtual deformations can be performed and the resulting haptic feedback can be felt. Thus, use of neuro-fuzzy technologies for the definition and calculation of virtual deformations seems applicable to the simulation of surgical interventions in virtual environments.