Yu.E. Gapanyuk - Ph.D. (Eng.), Associate Professor, Department of Information Processing and Control Systems, Bauman Moscow State Technical University E-mail: gapyu@bmstu.ru A.N. Nardid - Post-graduate Student, Department of Information Processing and Control Systems, Bauman Moscow State Technical University E-mail: anatnar@yandex.ru A.V. Gushcha - Post-graduate Student, Department of Information Processing and Control Systems, Bauman Moscow State Technical University E-mail: ncrashed@gmail.com

The article is devoted to the time series forecasting based on symbolic regression. Symbolic regression is a special case of genetic programming which means automatic program generation using genetic algoritms. With the help of genetic algorithm a program in form of syntax tree is generated using basic functions. Basic functions include exponential, logarithmic and trigonometric functions. The function values of generated syntax tree approximates the original time series. The fitness function of genetic algorithm is the approximation error. Genetic algoritm tries to reduce approximation error and to achieve maximum accuracy of the approximation. In the result of the analysis it was found the maximum number of individuals in the population, after which further increase of individuals does not improve prediction accuracy. The advantage of the proposed approach is the possibility of ad hoc learning of forecasting system.

 

1. Lehmann E.L. Symbolic regression via genetic programming. Casella, George: New York: Springer, 1998.
2. Koza J. Genetic Programming: On the Programming of Computers by Means of Natural Selection. United States: MIT Press, 1992. 609 p.
3. Carvalho L.. An Improved Evaluation of Kolmogorov-s Distribution // Journal of Statistical Software, Code Snippets. 2015. T. 65, № 3.
4. Snoyman M. Developing Web Applications with Haskell and Yesod. 1005. Gravenstein Highway North, Sebastopol, CA 95472: O-Relly Media Inc., 2012.