Karpenko

When solving practical, primarily, technical optimization tasks, the tasks of multi-objective optimization (MCHO-tasks) often arise, when it is desirable to exaggerate not one but several objective functions. It is well known that since, as a rule, it is impossible to achieve the extreme value of all objective functions simultaneously, the solution of such a problem is a certain compromise belonging to the Pareto set of the problem, that is, to a set of solutions that are not simultaneously improved for all objective functions. By the solution of the MCE-problem, we call the Pareto set of this problem, assuming that, after constructing some finite-dimensional approximation of the given set (P-approximation), the person making the decision (LPR), using informal or formalized methods, selects the unique solution from it. WEB-oriented software system PARETO-Q is designed for remote evaluation of Pareto-approximation quality in multi-purpose optimization problems. The system is designed to solve the problem of consolidation of Pareto-approximation algorithms and algorithms for estimating the quality of approximation. We represent the software used by the system, the architecture of the system, some results of computational experiments.

pp. 154-181

pp. 92-110

pp. 150-164

pp. 119-142

pp. 382-405

pp. 438-452

pp. 262-278

pp. 228-253

pp. 136-149

pp. 682-704

pp. 798-818

pp. 357-370

pp. 571-584

pp. 292-307

pp.181-193

pp. 193-214

pp. 215-242

The paper aims to present a developed mathematical model of the distribution urban power supply network taking into account its future development. Based on this model the task is set to optimize the prospective power supply network development through the solving a problem of multi-criteria structural and parametrical optimization. Expediency is proved to use a method of reduction to the one-criteria task, which represents a problem of continuous-discrete-integer programming to solve this task. Representation of this task is proved as a problem of discrete programming based on the discrete approximation of possible regions to construct new transforming and distributional substations.

The paper makes a statement of the multi-criteria optimization problem, gives a brief review of techniques to solve it using its Pareto-approximation, and presents mainly known approximation quality indicators. Principally, the paper gives a review of known programme systems orientated to the Pareto-approximation problem solution, which implement various algorithms to estimate the quality of this approximation.

The paper presents a canonical method of the swarm particles in two modifications to raise this method efficiency in solving multi-extreme problems of high dimension optimization. The method efficiency and its abovementioned modifications have been investigated in solving the test problems of optimization. One of the offered modifications has been used to solve the 28-dimensional molecular docking problem of proteases HIV1and darunaviry 3U7S as the molecules of receptor and a ligand, respectively.

The paper proposes an AIS_PSO algorithm (Artificial Immune System_ Particle Swarm Optimization), which represents a modification of the hybrid algorithm based on the CLONALG immune algorithm and the algorithm of the particles of swarm (PSO). The modification is to replace the CLONALG algorithm with the same, but the modified algorithm. The AIS_PSO algorithm has been realized by MatLab software package. An efficiency of developed algorithm and software has been examined using Rosenbrock's, Himmelblau’s, and Rastrigin’s test functions. The AIS_PSO software allows a problem solution for the optimal control of vibration robot movement. The paper applies a known method to solve the problem of the optimal control of dynamic system based on reduction of this task to a task of nonlinear programming of high dimension. The study results have shown that a developed algorithm and software are high efficient.

A task of evaluation of program's performance occurs often in the process of design of computer systems or during the iterative compilation. A traditional way to solve this problem is an emulation of program's execution on the target system. A modern alternative approach to evaluation of program's performance is based on the statistical modeling of program's performance on a computer under investigation. This statistical method of modeling programs' performance called Velocitas was introduced in this work. The method and its implementation in the Adaptor framework were presented. Investigation of the method's effectiveness showed high adequacy of program's performance prediction.

The authors propose a method for automated investigation of aorta areas affected by atherosclerosis, in order to find a relation between somatic mutations of the mitochondrial genome in the cells of aortic walls and a degree of manifestation of arterial sclerotic disease in these areas. This method consists of several basic steps; they are aorta image segmentation, marking of connected components of the image, plotting an adjacency graph of image areas and decomposition of this adjacency graph. A structure of software system was proposed; AORTA software system which implements the proposed method was designed and developed. Tests and approbation of the system in the SRI of general pathology and pathophysiology of RAMS demonstrated its working capacity and usability.

This work was carried out in the context of investigation and development of original two-component neutral catalyst systems made of mixed-valence metal compounds (titanium and zirconium) and organoaluminum compounds. A problem of identification of kinetic parameters of olefin hydroalumination, which belongs to a class of inverse problems of chemical kinetics, was considered in this paper. As a rule, identification problem is formulated as a single-criterion optimization problem for residuals between calculations and experimental data. The disadvantage of such a formulation consists in the fact that important prior information about kinetic features of a chemical reaction under investigation is ignored during identification. A distinct feature and novelty of this work is a new formulation of the identification problem; the problem was considered as a two-criterion optimization task which allowed to take into account both experimental data on the kinetics of the chemical reaction under investigation and prior information about this reaction. Solution to this problem represents the Pareto set. In order to approximate the Pareto set a modified Adaptive Weighted Sum method was used. One of the purposes of this work is to test the modified AWS method while solving real world two-criterion inverse problems of chemical kinetics. Two-criterion identification problem of kinetics of olefin hydroalumination by halbui2 and clalbui2 was formulated in this work. The algorithms and software which were used are described in this paper. The obtained calculation results were analyzed and discussed. 

This work is a part of investigation devoted to the development of design methods for ontological knowledge bases, oriented at the decision support in the enterprise information systems. An approach to finding solutions in knowledge bases using document’s metadata was considered. Document’s metadata as well as search queries were represented as the frames of design and search patterns respectively. Slots of those patterns corresponded to the concepts’ roles in the utilized ontology. Specified roles divided concepts of ontology, a document and a query into clusters. Semantic networks for those clusters were defined in such a way that search queries of a document were represented as a set of semantic networks, corresponding to the slots of design and search patterns. The relevancy of a document was estimated by a set of metrics which formalized the nearness of semantic networks. A problem formulation of multi-criteria estimation of the relevancy of documents in the enterprise ontological knowledge base and an adaptive solution method were presented in this paper.

This paper deals with the problem of multi-criterion optimization. It was assumed that the Pareto set is a solution to this problem. A promising approach to solving this problem is the adaptive weighted sum (AWS) method. Just like the classic weighted sum method based on the additive convolution of partial optimality criteria, the adaptive method also is based on the  convolution. But the AWS method also proposes adaptation of weighting coefficients during iterations on the basis of the current location of the search subarea. This method works with meta-models of criterion functions in order to reduce computational costs. Results of our investigations showed that the AWS method provides a high quality Pareto-approximation in case of a convex, although, maybe, unconnected Pareto frontier. For problems with a concave Pareto frontier this method doesn’t always provide sufficient quality of solution or it does but this takes too much time. In some cases the AWS method could provide an unacceptable solution caused by a special method of constraint satisfaction. The purpose of this work is to overcome specified disadvantages of the AWS method. A problem formulation for multi-criterion optimization was presented along with basic approaches to its solving. Several modifications of the ASW method proposed by the authors were considered. A brief description of the developed software implementing the method and its modifications were presented. Performance study of this software was carried out.

This paper deals with a co-evolutionary algorithm of global optimisation, Co-PSO, based on particle swarm optimisation. Implementation of this algorithm developed in MATLAB was described in this work. This software is oriented to a parallel operation of the given number of PSO algorithms at the logical level. These algorithms use various neighbourhood topologies of particles and/or various values of their free parameters. Results of a comprehensive numerical experiment on performance study of Co-PSO and its implementation were also presented. The experiment was based on Rosenbrock, Himmelblau and Rastrigin benchmark functions. Experimental results show superiority of the Co-PSO algorithm over the canonical PSO. A three-criterion problem of optimal control for a spacecraft during its reentry was solved with the use of Co-PSO. Methods of additive scalar convolution and reduction of the optimal control problem to a nonlinear programming problem were used in this work.

This article deals with a promising global optimization method that is a method of biogeography (BBO). Efficiency of this algorithm and its implementation was demonstrated by the example of the known complicated minimization problem of the multidimensional Rastrigin function. Integration of implementation of the biogeography method developed in MATLAB with PowerFactory software of DlgSilent was considered; PowerFactory software was used to calculate total power losses in the electric circuit being under consideration. Efficiency of the biogeography method for solving optimization problems of IEEE9 and Kuban’s power networks was investigated with the use of MATLAB BBO + PowerFactory. By optimizing a set of compensators and their physical layout, real-power losses in the specified networks were successfully reduced by 9.2% and 2.5% respectively.

In this paper we consider a global optimization problem and a method of Cuckoo Search (CS). CS method could be used for solving the specified problem. This method belongs to a new class of population based algorithms, which have been developed actively in recent years. In contrast with other known population based methods CS has only two free parameters. The goal of this work is to increase the efficiency of the canonical CS method. Several different modifications were presented in this work along with their software implementations; the canonical method was also implemented. Results of comprehensive performance study of proposed modifications by the example of a set of benchmark functions which revealed advantages over the canonical method were presented. One of proposed modifications was used for solving the existent practical problem of minimizing the expenditures on the manufacturing of high-pressure vessels. It was shown that proposed modifications provide a better value of the objective function in comparison with results obtained by other researches and also allow one to meet all restrictions.

A dodecapod is a one-section parallel manipulator which could be considered as an evolution of the wide-known hexapod, or the Stewart platform. The idea of a dodecapod was proposed by prof. S.N. Sayapin in 2012. In this article the authors consider one of the dodecapod’s functional capabilities: movement within a linear cylindrical tube of uniform and variable cross-sections. A design model of the dodecapod was presented in this paper along with propagation conditions for dodecapod’s movement within the tubes and the algorithms of changing the lengths of dodecapod’s rods, which make its movement possible. The presented results allow to draw a conclusion that a one-sectional parallel manipulator based on a dodecapod could be applied to creation of mechanisms for execution of technological operations in linear cylindrical tubes of uniform and variable cross-sections.

This is the second paper in the series of four works dedicated to design of parallel mechanisms for orientation of the space observatory “Millimetron”. In this article the authors stated a problem of “Millimetron” observatory's antenna orientation with the use of a parallel hexapod-based multi-sectional manipulator. The manipulator was designed in Solid Works software package. The Manipulator's 3D-model created in NX5 with the use of “Kinematic Simulation of Mechanisms” software package is introduced in this work. Software for solving direct and inverse kinematic problems was designed. A model of the guided five-section manipulator developed in MatLab Simulink software package is also described.

 The article deals with a three-objective optimization of slotted filter geometry produced by an innovative method of deforming cutting. This method was developed by Professor N.N. Zoubkov from BMSTU. The multiobjective problem is solved by using the interactive PREF method proposed by the authors. This method is based on the assumption that a decision maker (DM) is able to provide, in some form, preference information related to various solutions proposed by multiobjective optimization software. The DM’s utility function is approximated by a neural network. This work presents a three-objective optimization problem statement of filter design, a short description of an interactive method and its software implementation along with results of computational experiments. 

The authors consider an algorithm for solving systems of linear algebraic equations (SLAE) with preconditions. The authors introduce parallel algorithms and software for image processing devices that implement the basic operations of the algorithm - multiplication of the matrix by a set of vectors and solutions of a block triangular SLAE. The results of extensive studies of effectiveness of the proposed algorithmic and software solutions are provided in this paper. These results show a fairly high efficiency of the developed algorithms and software for multiplication of the matrix by a set of vectors. Acceleration of computations in this case is four to sixteen times. Algorithms and programs for solving the block triangular SLAE showed satisfactory results which allow to expect an acceptable acceleration for practically significant SLAE of high dimension and when using professional GPU.

The work deals with the problem of identification by voice. The focus is on building an adaptive classifier of specific voice features based on the Kohonen self-organizing map. The authors present the algorithms used for audio processing and classification, as well as the software system VOICE which implements these algorithms. The results of the studies confirming the effectiveness of the algorithmic and software solutions are given in this paper.

This publication continues the authors’ work in the field of ant algorithms of global continuous optimization. The authors propose a modification of the known HCIAC algorithm which was proposed by J.Dreo and P. Syarri in 2007. This modification is based on a set of memes in order to make local search more efficient. The authors present the results of their research on effectiveness of the modified algorithm - HCIAC-M – during the search of global extremum of Rosenbrock and Rastrigin functions. As an example, the authors consider practical importance of optimization of the engine-generator system.

The paper presents an original hybrid method of global optimization – SIA – which based on the artificial immune systems technology. The method represents extreme modification of the hybrid optimization method HIA (Hybrid Immune Algorithm) which uses mutation procedure for local search. On the contrary, in SIA method for this purpose the known SUBPLEX algorithm is used. The authors give a description of this method and its program implementation. In addition, the authors investigate efficiency of the proposed method and present the results of this research.

The authors consider the ontology knowledge bases focused on support of decision-making in corporate information systems. The authors believe that search for decision in these knowledge bases is carried out with the use of metadata of the document. Implementation of this approach leads to a graph-coarsening problem. The authors present some methods for solving this problem, and the results of the research of their efficiency.

The paper gives short descriptions of the particle swarm, mind evolution and clonal selection methods. Hybridization of the specified methods and a description of the MEPSI hybrid method (Mind Evolution, Particle Swarm, Immune) are presented. We consider an adjustment of free parameters of the HPSI method on the basis of which the MEPSI method is constructed. The results of studying an evolution of the swarm generated by the MEPSI method, as well as the results of studying the efficiency of the method for known Shekel and Rastrigin benchmark problems are presented in this article. Besides, the paper presents similar results obtained during solving the seven- and thirty eight Lennard-Jones atomic clusters problem.

The authors consider the problem of integral estimation of multi-criterion alternatives; they propose a formal statement of the problem and methods of solving it. The method transforms the problem into global conditional multidimensional optimization. To solve that problem Mind Evolutionary Computation algorithm is used. Implementation of this method is described. The results of efficiency research are given.

The authors consider solving the discrete problem of Pareto set and frontier optimization in the multi-criterion optimization problem. The aim of this work was to examine the efficiency of Adaptive Weighted Sum method (AWS) which was initially developed by J-H. Ryu, S. Kim and H. Wan. Comparing to the original research, the authors used a wider range of multi-criterion benchmark optimization problems. A range of limitations of AWS methodology and usage complexity was found. Modification of the method was proposed to resolve the problems and limitations.

This authors present a review of numerical methods of approximate Pareto set generation in the multi-objective optimization problem. The following methods are discussed: "naive" methods, switching objective functions methods, methods of objective functions aggregation, methods based on ranking of population agents, and other methods. All cases referred to methods involving the use of genetic or swarm algorithms, such as particle swarm optimization algorithm.

This article deals with co-algorithmic hybridization of two Particle Swarm Optimization (PSO) algorithms, using click-type and ring-type neighborhood topologies of particles. Each of these topologies has their disadvantages which can be eliminated by using co-algorithmic hybridization. The authors compared the efficiency of canonical PSO and its co-algorithmic modification and found optimal values of free parameters.

 This article describes an innovative mental-structured educational technology focused on training technical specialists, built on the basis of human memory and thinking laws.  This training provides technical specialists’ high competitiveness in rapidly changing market requirements for educational services and intellectual work. There are five basic principles of mental-structured educational technology. 

Dynamic system attainability domain approximate construction problem based on “multi-finish” and model differential equation system vector field approximation methods combination is discussed. The main point of the first method is multiple model differential equation system integration with different controls values. The idea of the second method is preliminary approximation  of model differential equation system’s right side. The purpose of this article is to find out the efficiency of neural network approximation. The research is carried out for 7^th order differential equation system, which describes aircraft center-of-mass motion. Three-layer feed-forward neural network and similar cascaded neural network are analyzed.

The research was a part of the project aimed at developing an automated optimal insulin dose forecasting system, which, when combined with an insulin pump and continuous glucose monitoring system, could serve as an artificial pancreas. Paper investigates the effectiveness of neural network algorithms for predicting the blood glucose values in type I diabetes patients. The aim is to choose the optimal neural network and its training algorithm for use in predicting optimal doses of insulin. This research was performed using a software system MatLab.

The paper presents the review of the basic mathematical models used in modern learning systems. We discuss the models of knowledge, the learner, the learning process, the control of educational achievements, and also some other models. As models of knowledge we present the frame-model, formal-logic models and models based on semantic networks. Such classes of the pupil models, as overlay models, imitating models, difference models, perturbation and stereotypical models are considered. From all variety of the training process models the review includes the models based on the Petry networks, probability models and the models based on the finite-state automations. As models of the control of educational achievements of the learner the paper presents the models based on the classical test theory (CTT) and models based on the item response theory (IRT).

Dynamic system attainability domain approximate construction problem is considered. We studyefficiency of using composition of multiple integration approach and vector field approximation of model differential equations system for solution of this problem. We present description of specified methods and their tests on application of approximate construction of attainability domain of aircraft, described by model. For given research we reach a conclusion about usability condition of approach composition involved.

The paper presents the general questions connected with a test quality monitoring of quality of training, and also various ways of classification of tests and test tasks. We present the review of the basic criteria for estimation of tests reliability, validity, difficulty and so on. A number of the validity measures of tests is offered. These measures are based on the knowledge of a subject domain of studied discipline in the form of a semantic network.

This paper discusses a problem of multi-dimensional continuous global constrained optimization and continuous interacting ant colony algorithm to solve this problem. The work presents the results of extensive research on the effectiveness of this algorithm for solving a number of test multi-extreme and ravine optimization problems. Recommendations about optimization of a choice of free parameters of algorithm have been made on the basis of results of research. This paper also discusses a problem of approximate construction of the Pareto set for multi-objective optimization.The effectiveness of combination algorithms CIAC and MOPSO is investigated to solve this problem.

The paper is dedicated to an approximation of a set of attainability for type “trunk” multisectional robot-manipulator. SPSO method was developed and used for the task resolution. SPSO is based on well-known MOPSO method that is used for multi-objective optimization task. SPSO method as well as MOPSO method implements one of the version of particle swarm optimization method (PSO) that is often used for global optimization tasks. Sigma-algorithm is used for finding non-dominated solutions. SPSO effectiveness is demonstrated for several tasks of a set of attainability search for both with and without obstacles.

The problem of parametrical optimization of search optimization algorithms (meta-optimisation problem) is considered. The meta-optimization method which basic feature consists in search of optimum strategy of algorithm in the course of normal operation of the program realizing it is offered. The method uses sets clusterization of the considered class problems by means of Kohonen network and the decision actually problems of meta-optimisation by means of continuous genetic algorithm.

Tasks of optimizing form and sections dimensions of multisectional «trunk» robot-manipulator are considered. Approach based on manipulator representation by a cantilever bar and three following criteria of optimality is used for solving the first task: criterion of uniform flexural resistance using beam of uniform strength theory; criterion of uniform specific power rating; criterion of uniform energy density. Criteria listed above are used for optimization of manipulator form under different types of load: static load, exerted by external forces and torques; static load exerted by beam’s weight; dynamic load caused by accelerated motion of beam’s components; dynamic load caused by head resistance of environment through which the beam is moving. Manipulator is also represented by a cantilever bar for solving the second task. Approach based on criterion of uniform section stiffness is used.

This paper presents multi-section trunk robot-manipulator based on parallel kinematics “hexapod” mechanism. The paper purpose is working out of mathematical model and dynamics research of this manipulator. Means of program system MatLab/Simulink develop dynamic models of one-section and three-section manipulators. Research of transients and the maximum loadings in their drives is executed at various weight of tong and parameters of automatic control system.

The paper discusses ontology knowledge bases focused on support of decision making in corporate information systems. We assume that search of decisions in knowledge bases is performed using metadata of the document. Metadata are formed on the basis of the semantic network of the ontology. Relevance of the document is estimated by computing affinity in metrics of the semantic network of the document and the semantic network of the query.

The article considers execution level of trunk-manipulator, considers control system of the tripod as a section of the manipulator. The article describes classical nonparametric identification and neural network identification based on NNARX model. Both method of identification were implemented in Matlab with Neural Network Toolbox and System Identification Toolbox. The paper considers classical  and neural network nonadaptive control system of the tripod. Classical two-circuit control system based on PID controllers. Neural network control system based on inverse model of the tripod. Also the article describes adaptive control system. There are two method describe in article. First method  based on classical control system with the reference model of tripod. Other method based on neural network control system. All neural networks model were implemented in Matlab with Neural Network Toolbox.  Modeling of control systems were implemented in Matlab/Simulink.

One-section and two-section manipulators based on parallel kinematics “hexapod” mechanism are considered. Linear and angular dimensions optimization problem is set. Minimal guaranteed suppleness of construction under applied force is used as criterion of optimality. The research is carried out by means of MatLab software system. MatLab program implementing SQP optimization method is used for solving the optimization problem. Mathematical models of manipulators are obtained by means of MatLab/Simulink system.

This paper presents the bees algorithm into global multidimensional optimization problem. We state biological foundations of the algorithm, shame of used method version and structure of corresponding Python program. We present also results of wide parametric investigation of method efficiency. The investigation carries by help of multidimensional test functions, which belong to three function sets - single-extremum functions, multi-extremum functions having not many extremum, multi-extremum functions having a lot of extremum.

This paper presents some results of our investigations in area of ontology knowledge base. These investigations have methods development for support decision-making into corporative information system as a goal. We use the models of semantic networks of ontology knowledge base and ontology knowledge a document of this base as the weighted multi-graphs. A number of these networks metrics is introducing. The metrics are base on topology behaviors of these graphs. On foundation of the metrics, we realize clusterization of semantic network of ontology knowledge base. Next, we define the weights of the multi-graph nodes and its arcs of this network. Finally, the paper presents a number of concept importance measures into semantic networks of knowledge base. The measures can use to estimate relevance of database queries.

A direct adaptive method of multicriteria optimization based on neuro-fuzzy approximation of decision maker’s function is considered in this paper. Some results of the investigation of the efficiency of the method for solving two- and three- criteria test problems are presented.

The article describes a parallel method of global unconditional optimization by particle swarm. Features of implementation in CUDA graphics processors are described. Study results of convergence, performance and scalability are given.

This paper presents the problem of goal configuration planning of a trunk robot-manipulator. We discus target setting, methods and C-  program for this problem. Also we present some results of goal configuration planning

Dynamic system attainability domain approximate construction problem is discussed. We analyze this problem solution approaches based onconsecutive and parallel “multi-finish” methods, respectivedifferential equations system vector field approximation, and neural nets. As an example, we discuss specified methods application for aircraft attainability domain boundary approximate construction.

The paper considers the direct adaptive method for multi-criteria optimization based on fuzzy approximation of decision maker’s utility function. We provide some results of method effectiveness research for solving two- and three-criteria test problems.

We discuss a “trunk” robot-manipulator. The paper presents kinematic and dynamic models of type «hexapod» parallel mechanism as a section of this robot-manipulator.

 We discuss a “trunk” robot-manipulator. The paper presents kinematic and dynamic models of type «thripod» parallel mechanism as a section of this robot-manipulator.

English language publications covering optimal insulin doses forecasting methods for insulin-dependant diabetes patients are presented in this review. Mathematical models of insulin dynamics in human body, neural network-based and composite algorithms and optimal insulin dose forecasting systems are considered.

This paper presents review of the particle swarm optimization method (PSO) for a global optimization problem and its certain serial and parallel modifications.

             The paper discusses evaluation of conceptual knowledge of a subject taught by intellectual learning computer system whose knowledge database is built on semantic networks. Cognitive maps are used to formalize the conceptual knowledge of a subject about the studied discipline.

The paper discusses a parameterized three-dimensional model of internal combustion engine mechanical subsystem as an element of computer aided system of engines.  The model consists of some units which makes it possible to build a model of engine with various construction schemes. The problem is solved by using software PRADIS for analysis of dynamic systems.

The review of criteria of similarity ontologies, and also display methods ontologies is resulted

 The paper introduces a concept “expanded semantic network of a tutoring system”. Also the paper discusses a number of complexity measures of this network - complexity measures for concept, shareable content object, object repository and content package 

The paper discusses a class of problems with two-level tree information structure graph with leaves’ computational complexity a priori unknown. Using LSF component of a well-known GRID-system Globus for distributed heterogeneous computing system load balancing while parallelizing problems of the specified class is described.

The paper discusses a number of complexity measures for concept, shareable content object, object repository and content package into a tutoring system