Control Systems and Computers

DOI: https://doi.org/10.15407/usim.2016.06.073

Upr. sist. maš., 2016, Issue 6 (266), pp. 73-79.

UDC 681.5.015

Moroz Olha G., junior research scientist, Departament for Technologies of Inductive Modelling, International Research and Training Center for Information Technologies and Systems of the NAS and MES of Ukraine, Glushkov ave., 40, Kyiv, 03187, Ukraine, Moroz.@ukr.net

Sorting-Out the GMDH Algorithm with the Genetic Search of Optimal Model

Introduction. Combinatorial GMDH algorithm (COMBI) is an effective means to solve the problems of structureparametric identification, forecasting, building models of objects and processes from experimental data under uncertain conditions.
But this algorithm is practically successful in solving such problems when the number of input variables (arguments) is less than 30 because of using exhaustive search of models with different structures from a given basic class of functions.
Partly this problem of exponential growth of the algorithm complexity is solved by some algorithms with directed models
search but they have their own limitations.

Purpose. The latest results concerning a sorting-out COMBI-GA hybrid algorithm with genetic search of optimal model
as an alternative to known algorithms based on the determinate search procedures is summarized.

Methods. The genetic-based algorithm uses the genetic operators to find an optimal model and reduce the exhaustive search of the combinatorial algorithm.

Results. The effectiveness of four genetic operators is compared in COMBI-GA hybrid algorithm for solving test and real-world inductive modeling problems of diverse dimension.

Conclusion. Using GA is effective way to search for optimal model in sorting-out GMDH algorithms for quick solving inductive modelling tasks with large numbers of input variables (much more than 30).

Download full text! (In Ukrainian and Russian)

Keywords: combinatorial GMDH algorithm (COMBI), genetic algorithm, hybrid GMDH-GA algorithm.

1. Stepashko V.S. Combinatorial program for the structural identification of objects and control processes. Ed. A.G. Ivakhnenko. Kiev: Tekhnika, 1980. pp.80–86.  (In Russian).
2. Stepashko V.S. Combinatorial algorithm GMDH with the optimal scheme of sorting models. Automation. 1981. 3. pp. 31-36.  (In Russian).
3. Ivakhnenko G.A. The use of predictive algorithms GMDH for solving marketing problems. Problems of management and computer science. 2008. 2. pp. 149-152.
4. Chaudhuri R.A. Combinatorial GMDH Approach to Identification, Modelling and Prediction of Money Market State Variables. Int. Conf. on Technology and Business Management, 2012. pp. 208–221.
5. Kondyrina A., Savchenko E., Direktorenko O. Forecasting of Ukraine Human Development Index by Combinatorial GMDH Algorithm. The 4th Int. Workshop on Inductive Modelling IWIM, 2011. pp. 46–50.
6. Samoilenko O. Combinatorial GMDH Algorithm Application
to Predict the Dynamic of Demographic Characteristics in Ukraine. 4th Int. Conf. in Inductive Modelling, ICIM’ 2013. pp. 312–314.
7. Savchenko, E.A. Express prognosis of glucose level in 7 blood according to the combinatorial algorithm of GMDH. Upravlausie sistemy i masiny. 2003. 2. pp. 107–112.
8. A.G. Ivakhnenko, E.A. Savchenko, G.A. Ivakhnenko et. al. Detection of patterns of interaction of ions with the surface according to the combinatorial algorithm of GMDH. Problems of management and computer science. 2003. 2. pp. 80–89.
9. Stepashko V.S. The final selection procedure for reducing the total enumeration of models. Automation. 1983. 4. pp. 84–88.
10. Samoilenko O.A., Stepashko V.S. Analysis of the effectiveness of the use of the frequency criterion in the algorithm of sequential screening of non-informative arguments. Inductive modeling of complex systems: Coll. sciences works. K .: MNNCIT and C NASU, 2012. pp. 191-209.
11. Efimenko C.M. Combined algorithm of MSUA with sequential complication of model structures based on recurrent-parallel computations. Inductive modeling of complex systems: Coll. sciences works. K .: MNNCIT and with NASU, 2014, 6, pp. 81-89.
12. Stepashko V., Moroz O. Hybrid Searching GMDH-GA Algorithm for Solving Inductive Modeling Tasks. The 1th IEEE Int. Conf. on Data Stream Mining & Processing. Lviv, Ukraine, 23–27 Aug., 2016. Lviv
Polytech. Nat. Univ., 2016, pp. 350–355.
13. Ivakhnenko, A.G., Stepashko, V.S., 1985. Noise-immunity of modeling. Kiev: Naukova dumka, 216 p. (In Russian).
14. Seber J. Linear Regression Analysis. M .: Mir, 1980. 456 p.
15. Hollannd J.H. Adaptation in natural and artificial systems. An introductory analysis with application to biology, control, and artificial intelligence. Univ. of Michigan, 1975. 210 p.
16. Gladkov LA, Kureichik VV, Kureichik VM Genetic Algorithms. M .: FIZMATLIT, 2006. 320 p.
17. Kaya Y., Uyar M., Tekin R. A Novel Crossover Operator for Genetic Algorithms: Ring Crossover. Presented at CoRR, 2011.
18.Glybovets M., M. Gulaeva N. M. Evolutionary algorithms: Tutorial. – K .: NaUKMA, 2013. 828 p. (In Ukrainian).
19. Moroz O.G. Efficiency of mutation operators in the task of genetic search of optimal models in the pre-selection algorithm of MSUA. Materials of the VIII Ukrainian-Polish sciences-practice. conf. “Electronics and Information Technologies” (ElIT-2016), Lviv-Chinadievo, August 27-30. 2016 Lviv: View. Center of LNU them. Yves Franco, 2016. pp. 331-332. (In Ukrainian).
21. http://archive.ics.uci.edu/ml/
22. https://archive.ics.uci.edu/ml/datasets/SkillCraft1+Master
+Table+Dataset

 Received 23.11.2016