|
|
 Выпуск 95
- Glushchenko A., Lastochkin K., Petrov V. Procedure of identification of piecewise-constant unknown parameters with improved convergence
The research is aimed at improvement of the solution quality of the unknown piecewise-constant parameters identification problem for the classical linear regression equation. To solve this problem, a new procedure to process such equation, which is based on the known method of integral dynamic extension and mixing (I-DREM) but with the interval-based integral filter with exponential forgetting and resetting, is proposed. As proved in the paper, when the I-DREM procedure is applied, the proposed filter, unlike known from the literature, allows one to generate the regression equation with a scalar regressor and adjustable level of disturbance, which is caused by the step-like change of the unknown parameters. The main result of the study is a procedure to process a linear regression equation with a vector regressor, which allows one to derive an adaptation law. If the condition of the regressor finite excitation is met, then such a law guarantees that the identification error of the piecewise-constant parameters is bounded by an adjustable value. All of the aforementioned properties are proved analytically and/or demonstrated via the numerical experiments.
- Layko A. Model of structure optimization developing organizational system
The problems of finding optimal control in the extended game-theoretic models "Center-Agents" and "Center-Manager-Agents" are investigated, which consist in analyzing the quality of control with an increase in the number of elements, assessing the maximum load on the subject of control, identifying the need to add an intermediate link and determining the benefits its availability for the top-level The concept of a criterion for the effectiveness of management has been introduced and its changes have been recorded in connection with an increase in the elemental composition of the organizational structure. In order to determine the feasibility of introducing an intermediate managing subject, specific examples of generalized mathematical models of organizational systems are considered, their numerical solutions are presented, the dependences of the optimal control parameters on the number of employees of the organization are found.
- Grishin E. Approximate aproach in proteomics
Modern mathematical methods for protein analysis, such as database search and de novo methods, have their own drawbacks. It is not possible to identify proteins that are not included in databases using database search. The de novo methods allow us to identify new proteins but they are very computationally demanding (requiring the use of a supercomputer). In this project a complex approach of approximate protein analysis conducted on a personal computer was developed. A problem of qualitative and quantitative determination of initial sequence (protein) consists of three subproblems. The first one is noise cancellation and peak identification using mass spectrometry data. An algorithm combining a sliding average method and computational photography HDR technology was developed. The second subproblem is peak identification. It was reduced to a knapsack problem and solved using the branch and bound method. The last subproblem is initial sequence reconstruction using a set of fragments (peaks and their intensities). This subproblem was solved by constructing double trees and searching for a path of maximum length. All calculations were performed on a PC using CUDA parallel computing technology.
- Fomin A., Glushchenko A., Poleshchenko D., Zorin I. Identification of steel billets marking at rolling shop No. 1 of JSC "A.A. Ugarov Oskol Electrometallurgical Plant" using neural networks
The report is about the solution of a steel billet identification problem before its loading into the methodical furnace of a rolling shop of a metallurgical plant. The task is to develop an automated system that allows one to relieve the loading control station operator from the task of manual identification of each billet. The article considers such approaches to solve the problem in question as application of additional markings to the billets for the purpose of further automatic identification, and development of a system for recognition of the existing markings, which is based on the neural networks. An experiment on marking of the "gray" and "light" billets is conducted with the help of the laser markers of different power. The readability of the obtained codes is evaluated. The conclusion is made that laser marking can be applied only on clean rolling metal (without oxide scale). Therefore, the method is proposed to develop a neural network-based system for identification of the existing marking and implement it. The system is successfully introduced into production and allows one to achieve approximately 90% recognition accuracy, which reduced the operator's workload and the probability of steel grades "mixing" inside the heating furnace.
- Zhukov P., Glushchenko A., Fomin A. Comparison of finite difference and data-based models application to solve problem of steel billets temperature prediction
Within the whole cycle of technological conversion processes, which are widespread in the ferrous metallurgy, there are many energy-intensive technological units, energy-efficient control of which is a complicated task due to the non-stationarity of technological processes within them. One of such units is a continuous flame furnace, which is used for simple heating, homogenization, annealing and other operations. If the surface temperature of the billets at the furnace outlet could be known in advance (while they are still inside the furnace), it would be possible to adjust certain heating parameters, while staying within the technological instructions, in order to minimize the consumption of the combustible fuel. Therefore, in this paper we compare two models to predict the temperature of the billet surface after a simple heating in such a furnace: a model based on numerical differentiation of transient heat conduction equation and a tree-like one, which is obtained by machine learning and based on the technological data from the lower level of the furnace automation. It is supposed that such models can become the basis for a "digital twin" of the unit, which can be further used in Advanced Process Control (APC) systems. As a result of comparison, it is obtained that the error of the data-based model is 7.4 degrees Celsius lower on average comparing to the finite-difference one. It is assumed that this result is a consequence of the advantage of the first model "natural adaptation" to the technological unit.
- Kokunko J. Designing a differentiator for a control system for a mobile robot
In the framework of the cascade approach to state observers design for dynamic objects under the influence of external uncontrollable disturbances, a method for or reconstructing the derivatives of any desired order of a deterministic time function from its current values, which does not require knowledge of the analytical form of the function and numerical differentiation, is proposed. Assuming that the function is piecewise smooth and its derivatives are bounded by known constants, a virtual dynamic model of canonical form with an unknown input is introduced. On the basis of this model, whose order depends on the order of the derivatives to be recovered, a dynamic differentiator is constructed in the form of a state observer with piecewise linear corrective actions. In this paper, the above designs are demonstrated on the example of a control system for a wheeled robot. An nonlinear control law that stabilizes the motion of the platform along an admissible curvilinear trajectory is used to synthesize the feedback. The current information about the state variables of the control plant model, the setting influences and their first- and second-order derivatives is required to implement the feedback in the problem of track stabilization.
- Tkachenko A. Design of MPC-based control system for stabilizing a two-wheeled self-balancing robot
The purpose of this work is solving the problem of designing a control system based on Model Predictive Control (MPC) for a linear Multiple Input Multiple Output (MIMO) system of a two-wheeled self-balancing robot. These robots have become popular projects of many companies nowadays. One of the most famous companies engaged in the development of humanoid robots Boston Dynamics among them. Nevertheless, the idea of creating such type of robots did not pass by amateur engineers from all over the world, who began develop two-wheeled self-balancing robot for use as office assistants, vehicles and just for academic interest. The paper considers the process of constructing a mathematical model of the robot's mechanical system and linearization of the resulting model is performed. The basic principles of constructing an MPC-based control system for linear systems without external disturbances, as well as using an observer to evaluate the states of the model under the influence of additive white Gaussian noise are presented. Examples of implementation in the Matlab/Simulink environment for the case of a MIMO system, which is a model of a two-wheeled robot, are given.
- Blinova O. Search for optimal locations of network devices for wireless networks, taking into account the trajectories of the abonents and signal propagation
The use of wireless networks as a temporary solution for providing communication to participants of a number of works or events is complicated by the complexity and high cost of designing such networks. But the lack of communication in extreme conditions can lead to the death of people even if other resources are available to ensure security. There are network devices and equipment that allow providing wireless communication even in the most inaccessible areas, but for their effective use, an analysis of the feasibility of their location in specific places and the ability to assess the performance of the network as a whole is needed. The existing computer-aided design tools are focused on creating indoor networks and do not have sufficient funds for designing networks in conditions of a shortage of network equipment. The system discussed in the report is designed to simplify and accelerate the design of wireless networks in open areas with complex terrain. The problem of providing communication to subscribers traveling along specified routes with a limited set of network equipment is solved by searching for optimal locations and analyzing the provided communication level for each specified route point. An algorithm based on gradient descent and two phases of searching for the optimal location of devices are used. The simulation results are displayed graphically on a three-dimensional relief, as well as in a tabular form compatible with tabular processors for further processing. A module for digitizing terrain based on topographic maps or diagrams has been created.
|
|
