Issue 182, article 3
DOI:https://doi.org/10.15407/kvt182.02.025
Kibern. vyčisl. teh., 2015, Issue 182, pp.
International Research and Training Center for Information Technologies and Systems of National Academy of Sciences of Ukraine and Ministry of Education and Science of Ukraine
APPROACH TO ORGANOZATION OF RECURRENTAND-PARALLEL COMPUTATIONS IN AUTOMIZED STRUCTURE-PARAMETRIC IDENTIFICATION SYSTEM
Introduction. Development and optimization of methods and algorithms for solving statistical modeling problems are the basic directions in science undoubtedly. Although even the most efficient methods and technologies lose their value if they stay just program modules which can be used only by programmers. To be used in practice they should be integrated in some software that has intuitive user-friendly interface. Such software helps to discover a real value of the methods behind it.
The purpose of the paper is increasing the usage effectiveness of а recurrentand-parallel iterative algorithm by developing a full-fledged modern software based on it for solving forecasting, extrapolation and approximation problems.
Results. The main task of integration of GUI and computational module is a union of two mechanisms of parallization: Threading Building Blocks (TBB) parallization and Qt-parallization. The main idea of proposed approach is that every operation (including the model building operation that create own TBB-threads) initiated by a user should perform in additional Qt-thread. A design pattern that solve this task was developed. The pattern was used to finally implement the ASIS. The system applied to forecast investments security of Ukraine. A system of forecasting models that describe the state of Ukrainian investments security was build. Mean absolute percentage error of the models hit the [-7; +7] interval on independent dataset, that indicate a good forecasting ability of the models.
Conclusion. The work suggests an approach to organization of recurrent-andparallel computations in ASIS that allow integrating the most effective methods for solving statistical modeling problems in user-friendly intuitive full-fledge system that allows any user to solve forecasting, regression and approximation problems with high efficiency. The system was applied to forecast Ukrainian investments security. The economic interpretation of the obtained forecasts says that in 2013 year Ukraine will gain more investments from abroad than from inner investors Ukrainian manufacturing will increase rates; on the background rise of overall country investments, the investments in basic capital will decline.
Keywords: Multithreaded parallelization, design patterns, group method of data handling, recurrent-and-parallel computations, Qt, TBB.
References
- Efimenko S., Stepashko V. S, Basics of recurrent-and-parallel computations in combinatorial algorithm COMBI of GMDH // Controll sysmems and machines — 2014. — № 6. — P. 27–33. (in Russian)
- Efimenko S.M. Combinatorial algorithm of GMDH with sequential complication of structures based on recurrent-and-parallel computations // Inductive modeling of complex systems — Vol. 6. — 2014. — P. 81–89. (in Ukrainian)
- Pavlov A.V. Parallel relaxational iterative algorithm of GMDH // Inductive modeling of complex systems — Vol. 6. — 2014. — P. 33–40.
- Pavlov A.V. Design of automated structure-parametric identification system // Inductive modeling of complex systems — Vol.7— 2015. — P. 33–40.
- Internet resource https://ru.wikipedia.org/wiki/Qt.
Received 28.10.2015