Key words: approximation, time series, genetic algorithm, steady-state selection, chromosome representation, wavelets
Sadržaj: Problem optimiranja višedimenzijskih funkcija cilja javio se u mnogim područjima znanosti. Isto tako, problem određivanja koeficijenata aproksimacijskih funkcija može se svesti na problem optimiranja. Ugrađen je eliminacijski genetski algoritam s dva različita prikaza: binarni i prikaz brojem s pomičnom točkom. Opisani su uobičajni i posebno definirani genetski operatori. Predstavljan je genetski algoritam s uobičajnim genetskim operatorima nad brojevima s pomičnom točkom. Genetski algoritam je uporabljen za određivanje koeficijenata nekih aproksimacijskih funkcija: polinoma, trigonometrijskih funkcija i kvadratnih valića. Postignuti su dobri rezultati pri aproksimaciji vremenskih nizova s pomoću sinusnih funkcija i kvadratnih valića.
Summary: The problem of optimisation of multi-dimensional cost functions appears in many scientific fields. The problem of determination of approximation function coefficients can also be transformed into the optimisation problem. The implemented genetic algorithm has steady-state reproduction and uses two chromosome representations: binary and floating-point. Traditional and special genetic operators are considered in this paper. The floating-point chromosome representation with traditional operators is presented. For determination of coefficients of some approximation functions the genetic algorithm was applied. The polynoms, trigonometric functions and Haar wavelets were used. Very good results in time series approximation with trigonometric functions and wavelets by the genetic algorithm were achieved.