Software & Systems

Articles of journal № 3 at 2013 year.

Order result by: Public date | Title | Authors

41. Russian regions activity indicators analysis software system based on ontological models and data patterns [№3 за 2013 год]
Authors: Anton D. Moscowsky, M.A. Rovbo, Хорошевский В.Ф.
Visitors: 10539
An analyst workbench development and implementation based on intelligent mining of large amounts of statis-tical data in the domains of science, education and innovation is discussed in the paper. Scientific and technical challenges are pointed out. The R&D objectives are carried out: a specification of an indicators system of science, educationand innova-tion, as well as aggregation of statistical data for providing analysis of scientific, educational and innovation activity of Rus-sian regions in various aspects. Methods to form vectors ofspecific indicators and aggregates for statistical datasets are de-veloped. Business tasks are solved including identification of semantically similar vectors for building clusters of “similar” regions, and dynamic analysis of statistical data for monitoring performance of Russian regions in science, education and in-novation. A brief overview of the art state in domain is provided. A hybrid approach proposed in the paper integrates classi-cal mathematical methods for correlation analysis, pattern and time series analysis with methods of their semantic interpreta-tion. The developed software provides analytical support for trends identification, atypical dynamics of indicators and “best performance” patterns specification. Implementation of user-friendly interfaces is also discussed.

42. The prototype of the computer clusters state analysis system based on support vectors method [№3 за 2013 год]
Authors: Конюхов С.С., Московский А.А., Рябинкин Е.А., Велихов В.Е.
Visitors: 7792
Increasing complexity of computer cluster systems highlights the problem of failure tolerance to provide their efficient work. Nowadays this problem is one of the most dynamically elaborating in the area of high performance computa-tions. The paper considers different implementation aspects of automated system performing classification of computer clus-ter states to predict and check critical events. The algorithm using statistical learning theory approach (support vector meth-od) is proposed to address these subjects. Its modular structure permits to use more robust and efficient methods to deal with subtasks of the general algorithm. The the algorithm implementation prototype is tested on sets of real-world data. Provided that correct parameters are selected its anomaly detection capability is shown. In the test experiments the prediction error of future possible states is 6–16 % while the critical change of control parameters is fixed within 3–5 prognosis time steps. Sen-sor failure tolerance with small numbers (5–10 %) of data series with outliers is proved. Possible improvements of the auto-mated classification system are related with on-line SVR algorithms and methods to deal with more sophisticated anomalous behaviour of computer clusters.

43. Data model developing of information system to support lifecycle of aerospace electronic devices [№3 за 2013 год]
Author: Вичугова А.А.
Visitors: 9624
The article describes the concept of design objects, their correlation and the component structure in relation to the space instrumentation problems. Information systems and technologies to support the processes of product design are considered. The scheme of diverse information entities structured storage in the data management system is shown. The as-pects of the conceptual design of a information system data model to support the product life cycle, including integration with CAD systems are supplied. The article conseders the need of electrical and mechanical engineering equipment, the classes of data models allowing describing the composition and the relationship of design objects. The UML-diagram that illustrates the hierarchy of object types and relationship of design as anexample of electronic equipment is developed. The example of de-veloped model applying to the design data storage in PLM-system Enovia SmarTeam is described.

44. Expert systems design based on the transformation of domain information models [№3 за 2013 год]
Authors: Grishchenko M.A., Yurin A.Yu., Pavlov A.I.
Visitors: 11955
The paper describes the software to design rule-based expert systems. The software includes MDA methodolo-gy (Model Driven Architecture) and is intended for non-programmers. MDA is based on the stepwise transformation of do-main models. The proposed methodology feature is using ontology as a computation independent model (CIM); RVML (Rule Visual Modeling Language) as a platform independent model (PIM); JESS as a platform specific model (PSM). The architecture, main modules, functions and the universal logic data model for storing and interacting ontology and rules are presented. The RVML (designed to describe the rules with theaid of graphical primitives) is described. RVML increases the efficiency of rule-bases development by improving the clarity and comprehensibility of visualization of cause-effect relations (as compared with UML).

45. A distributed platform for parallel training of DisANN artificial neural networksA distributed platform for parallel training of disann artificial neural networks [№3 за 2013 год]
Authors: Нгуен Занг, Краснощеков А.А.
Visitors: 11483
The article describes the process of design, development and implementation of distributed neural network learning system based on learning-by-block algorithm for feed-forward neural networks. Distributing occurs in the array of training vectors which is divided into several blocks, transmitted to computing nodes. Thus, the neural network training is parallelized within the same epoch. Since neural networks are resistant to errors, losses are not critical. To avoid synchroniza tion bottlenecks in the end of each epoch, training may be carried out with losses of blocks. When synchronizing and transit-ing to the next epoch only a certain percentage of used training blocks or predetermined processor time is required. Within each period loss in training is compensated with the weighted value of last successful result. The system is based on a previ-ously created Anthill open-source platform for grid-computing. Applying the concept of grid-computing, the ability of con-trolling through web-interface and free choice of libraries for neural networks simulation allowed creating a system with a high degree of flexibility and ease of use.

46. The distributed subsystem of the electronic circuits design [№3 за 2013 год]
Authors: Glushan V.M., Лаврик П.В.
Visitors: 10756
The chronological express analysis of the approaches to constructing high-speed CAD electronic schemes is given. The attention is directed to the fact that the systems of automated designing should satisfy the interactivity require-ment. This requirement can be realized when the system gives replies to the user inquiries with a delay not exceeding2-3 se-conds. In the conditions of constant increasing of projected objects complexity the maintenance of automated designing sys-tems interactivity is impossible without constant escalating of their high-speed properties. The article considersthree essen-tial directions of CAD interactivity maintenance. Taking into account general expansion of network technologies, using the potential of various kinds of networks for creation of distributed CAD is a perspective and promising direction of researches. There is the structure developed by authors and the results of imitating modeling of distributed CAD that substantiated the expediency of constructing a real subsystem of designer projecting of electronic schemes. The basic modules of the subsys-tem, their functioning and results of experimental researches are described. Experimental researches have shown that the de-veloped subsystem is referred to the type of GRID-systems and allows reducing time of projecting to 3 times.

47. Realization of intelligent systems of real time on the basis of the Petri nets with support of temporal dependences [№3 за 2013 год]
Authors: Eremeev, A.P. , Королев Ю.И.
Visitors: 7370
The apparatus of color Petri nets in respect of its use in intelligent systems of real time is considered. Petri nets allow to describe naturally synchronization, overlapping, the conflict and causal dependence, and also visually to represent a structure and behavior of difficult technical or organizational systems. However, an essential drawback of classical Petri nets is complexity of modeling real time systems when it is necessary to consider a factor of time and temporal dependences. New expansion – color Petri nets with support of Allen’s temporal logic, allowing to model both quantitative, and qualitative temporary delays is offered. This logic is characterized by sufficient expressiveness and existence of polynomial algorithms of a conclusion that its practical application in intelligent systems like intelligent decision support systems of real time allows. The description and the example of use of the developed basic software for this class of Petri nets is given.

48. Solving average delay cost routing problem [№3 за 2013 год]
Authors: Margolis B.I., Muzanna M.M.
Visitors: 9217
The article describes a flow model of an information network with alternative routing of messages to any num-ber of switching nodes and links. The authors formulate the optimization criteria for the delay posts, restrictions on the сhannel сapacity and the conservation of flows on the network. Analytical relations for information flows optimal distribu-tion in a network of three lines are recieved using the method of Lagrange multipliers. Obtained relations are extended to var-ious numbers of lines. The paper shows formulas validity tocalculate the optimal flow distribution for the specialcase of a network consisting of two lines. The calculation program inMatLab is developed. It allows for the topological network struc-ture, the input streams matrix and bandwidth to find the optimal flow distribution in the network consisting of an arbitrary number of nodes and links. The article shows an example of the input flows optimal division for a network of four nodes and three lines, providing a minimum average latency. Changing the flow optimal distribution for this example is shown with a decrease in the number of lines to two. The paper outlines approaches to the problem of flows optimal distribution with in-termediate nodes in a simulated network topology.

49. Solution of a software structural construction problem for intelligent humidity sensor [№3 за 2013 год]
Author: Артемова С.В.
Visitors: 8215
The article considers the approach to solution of softwarestructural construction problem using the intellectual humidity sensor (IHS), which is an information-measuring system (IMS).When designing software and data support, the main modules of HIS are: an intellectual interface, a supervisor, survey sensors modules and measurement procedures mod-ules. The functioning of the IHS is based on using databases and knowledge bases. The IHS implementation requires a num-ber of algorithms, which should be placed in the modules to provide the designed IMS with specified properties. Such tasks of a choice belong to the class of NP-complete that can be solved only approximately. Therefore, for practical solving of the software structural construction problem it is offered to use heuristic methods, in particular, genetic algorithms. Method to solve the tasks includes following stages: forming the vector-line of algorithms distribution genes on the modules,description of a matrix of relations, interaction of algorithms using information elements, creating the matrix of alternatives ofalgorithm belonging to module, determining the target function. This approach can be used for designing data support and software for information-measuring or information management systems, that involves a large number of modules and algorithms.

50. A system for automatic traffic signs mapping [№3 за 2013 год]
Authors: Чигорин А.А., Konushin A.S.
Visitors: 17586
The article describes a system for the large-scale automatic traffic signs recognition and mapping. To apply au-tomatic methods successfully we need the cumulative time ofautomatic algorithm and time for error correction to be less than the time for manual data labeling by the human operator. Our system consists of four modules: detection, recognition, linking and localization. Detection and recognition mean that traffic sign position is determined on each frame separately and class label is assigned for each detected sign. On the linkage step the same physical signs from different frames arelinked in-to the sequence that allows refining detection and recognition results. Coordinates of the linked sequence aredetermined on the localization step by triangulation. We experimentally justifyed design choices made for different components of the sys-tem. The system works with more than 140 different classes of traffic signs and does not require labor-intensive labeling of large amount of training data due to the training on synthetically generated images. We evaluated our system on the large da-taset of Russian traffic signs and made this dataset publicly available to encourage future comparison.

◄ ← Preview | 1 | 2 | 3 | 4 | 5 | 6 | Next → ►